{ "$schema": "https://modeldeprecations.dev/api/v1/schema.json", "generatedAt": "2026-07-30T19:06:26.294Z", "asOf": "2026-07-30", "count": 156, "models": [ { "provider": "anthropic", "model": "claude-1.0", "name": "Claude 1.0", "aliases": [], "description": "The first Claude released to developers, in March 2023, and the model that introduced Constitutional AI to a public API. It used the text-completion endpoint with literal \"Human:\" and \"Assistant:\" turn markers rather than a structured message list — a shape that reads as archaeology now.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-1.1", "name": "Claude 1.1", "aliases": [], "description": "A mid-2023 refresh of the original Claude, largely a reliability and instruction-following pass rather than a capability jump. Anthropic shipped 1.0 through 1.3 within months of each other and then retired all four on a single day.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-1.2", "name": "Claude 1.2", "aliases": [], "description": "The third Claude 1 point release, and the one many early integrations pinned because it was the current default when the API opened more widely. Retired November 2024 with the rest of the line.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-1.3", "name": "Claude 1.3", "aliases": [], "description": "The last and most capable Claude 1, and the version most benchmark papers from 2023 actually measured. It was superseded by Claude 2 within a few months but stayed callable for over a year afterwards.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-2.0", "name": "Claude 2", "aliases": [], "description": "The first Claude available without a sales conversation, and the model that introduced most developers to Anthropic. It used the Human/Assistant text format rather than structured messages — a shape that has not existed in the API since. Retired July 2025.", "released_on": "2023-07-11", "deprecated_on": "2025-01-21", "shutdown_on": "2025-07-21", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-opus-4-8", "recommended": true, "note": "This is an API migration, not a model swap: prompts move from a single Human/Assistant string to a messages array with a separate system field.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-2.1", "name": "Claude 2.1", "aliases": [], "description": "The model that took Claude to a 200k context window in late 2023, when the rest of the field was at 8k to 32k. Long context was Anthropic's clearest differentiator for most of a year. Retired July 2025.", "released_on": "2023-11-21", "deprecated_on": "2025-01-21", "shutdown_on": "2025-07-21", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-opus-4-8", "recommended": true, "note": "Claude 2.1 predates the Messages API. Migrating means moving off the legacy completions endpoint and its Human/Assistant prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-5-haiku-20241022", "name": "Claude Haiku 3.5", "aliases": [ "claude-3-5-haiku-latest" ], "description": "The fast, cheap Claude that handled classification and routing work for most of 2025, and the last Haiku before the 4.5 generation. Retired February 2026.", "released_on": "2024-10-22", "deprecated_on": "2025-12-19", "shutdown_on": "2026-02-19", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Haiku 4.5 is substantially more capable at a similar price point, and supports extended thinking, which Haiku 3.5 did not.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-5-sonnet-20240620", "name": "Claude Sonnet 3.5", "aliases": [], "description": "The original Sonnet 3.5, from June 2024, and the model that introduced Artifacts. Anthropic later shipped a second model under the same 3.5 name, which is why anyone pinned to a date needs to check which one they meant. Retired October 2025.", "released_on": "2024-06-20", "deprecated_on": "2025-08-13", "shutdown_on": "2025-10-28", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-sonnet-4-6", "recommended": true, "note": "Anthropic points both Sonnet 3.5 snapshots at the same successor, so the June/October distinction stops mattering after the migration.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-5-sonnet-20241022", "name": "Claude Sonnet 3.5 v2", "aliases": [ "claude-3-5-sonnet-latest" ], "description": "The October 2024 Sonnet — confusingly the second model to carry the 3.5 name, which is why the dated id matters here more than anywhere else in the Claude lineup. It introduced computer use and was, for months, the model developers preferred over anything else on the market. Retired October 2025.", "released_on": "2024-10-22", "deprecated_on": "2025-08-13", "shutdown_on": "2025-10-28", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-sonnet-4-6", "recommended": true, "note": "The computer-use tool this model introduced has changed shape since; re-read the tool docs rather than porting the 2024 schema.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-7-sonnet-20250219", "name": "Claude Sonnet 3.7", "aliases": [ "claude-3-7-sonnet-latest" ], "description": "The first Claude with extended thinking, and the release that made a visible reasoning budget a mainstream API feature. It was retired in February 2026, almost exactly a year after launch.", "released_on": "2025-02-19", "deprecated_on": "2025-10-28", "shutdown_on": "2026-02-19", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-sonnet-4-6", "recommended": true, "note": "Sonnet 4.6 uses adaptive thinking rather than an explicit budget, so the thinking.budget_tokens field Sonnet 3.7 introduced no longer applies.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-haiku-20240307", "name": "Claude Haiku 3", "aliases": [], "description": "The cheapest Claude ever shipped, and the reason a lot of high-volume pipelines chose Anthropic at all. It outlived Haiku 3.5 by two months — Anthropic retired the newer model first, which is rarer than it sounds and caught out anyone who had migrated forward.", "released_on": "2024-03-07", "deprecated_on": "2026-02-19", "shutdown_on": "2026-04-20", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Haiku 4.5 costs more per token than Haiku 3 did. For pipelines chosen on price alone, re-run the unit economics before assuming this is a drop-in.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-opus-20240229", "name": "Claude Opus 3", "aliases": [], "description": "The first Claude to beat GPT-4 on published benchmarks, and the model that put Anthropic in serious contention. It had an unusually long deprecation window — six months — and its retirement prompted Anthropic's public commitments on preserving model weights after shutdown.", "released_on": "2024-02-29", "deprecated_on": "2025-06-30", "shutdown_on": "2026-01-05", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-opus-4-8", "recommended": true, "note": "Opus 4.8 is cheaper per token than Opus 3 was, has a 1M-token context window against 200k, and rejects temperature, top_p and top_k.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-3-sonnet-20240229", "name": "Claude Sonnet 3", "aliases": [], "description": "The middle model of the original Claude 3 family, positioned between Haiku and Opus on the price ladder that every provider has copied since. Retired July 2025.", "released_on": "2024-02-29", "deprecated_on": "2025-01-21", "shutdown_on": "2025-07-21", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-sonnet-4-6", "recommended": true, "note": "Two full generations separate these models. Prompts written to compensate for Sonnet 3's weaknesses usually hurt rather than help on 4.6.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-fable-5", "name": "Claude Fable 5", "aliases": [], "description": "Anthropic's most capable widely released model, built for agents that run for hours rather than turns. Anthropic publishes a \"not sooner than\" date for every active model, which is unusual candour: you know the floor on a model's life the day it ships.", "released_on": "2026-06-09", "earliest_shutdown_on": "2027-06-09", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "name": "Claude Haiku 4.5", "aliases": [ "claude-haiku-4-5" ], "description": "The fastest Claude with near-frontier quality, and the named replacement for every retired Haiku and Instant model going back to 2024. Active, with an earliest retirement date of October 2026.", "released_on": "2025-10-01", "earliest_shutdown_on": "2026-10-15", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-instant-1.0", "name": "Claude Instant 1.0", "aliases": [], "description": "The first of Anthropic's cheap, fast models — the ancestor of everything now called Haiku. Instant traded depth for latency and price, which is what made Claude viable for classification and moderation work in 2023.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-instant-1.1", "name": "Claude Instant 1.1", "aliases": [], "description": "The middle Instant release, and the one that widened the context window enough to make document summarisation practical at Instant pricing. Retired alongside the whole Claude 1 generation in November 2024.", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Two generations and a whole API separate these. Claude 1 predates the Messages API, so a migration means moving off /v1/complete and its \"\\n\\nHuman:\"/\"\\n\\nAssistant:\" prompt format entirely.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-instant-1.2", "name": "Claude Instant 1.2", "aliases": [], "description": "The last of the Instant line — Anthropic's cheap, fast tier before Haiku existed. It was retired in November 2024 with two months notice, and the Haiku models inherited its role.", "released_on": "2023-08-09", "deprecated_on": "2024-09-04", "shutdown_on": "2024-11-06", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-haiku-4-5-20251001", "recommended": true, "note": "Haiku 4.5 is the modern equivalent of this tier. Like Claude 2, Instant predates the Messages API, so the request format changes too.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-opus-4-1-20250805", "name": "Claude Opus 4.1", "aliases": [ "claude-opus-4-1" ], "description": "The August 2025 Opus, and the last Anthropic model priced at $15/$75 per million tokens — three times what Opus costs today. It is the only deprecated Anthropic model still answering requests, and it has days left.", "released_on": "2025-08-05", "deprecated_on": "2026-06-05", "shutdown_on": "2026-08-05", "status": "deprecated", "replacements": [ { "provider": "anthropic", "model": "claude-opus-4-8", "recommended": true, "note": "Opus 4.7 and later reject temperature, top_p and top_k with a 400 error when set to a non-default value. Remove them from the request rather than tuning them down. Opus 4.8 also defaults effort to high on every surface, so set it explicitly if you were relying on a cheaper default.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "anthropic", "model": "claude-opus-4-20250514", "name": "Claude Opus 4", "aliases": [ "claude-opus-4-0" ], "description": "The Opus that launched the Claude 4 generation and introduced extended thinking with tool use. It was retired in June 2026 on two months notice, alongside Sonnet 4 — the first time Anthropic retired an entire generation in one announcement.", "released_on": "2025-05-14", "deprecated_on": "2026-04-14", "shutdown_on": "2026-06-15", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-opus-4-8", "recommended": true, "note": "Opus 4.7 and later reject temperature, top_p and top_k with a 400 error when set to a non-default value. Remove them from the request rather than tuning them down. Extended thinking has also been replaced by adaptive thinking: thinking.type: enabled is no longer accepted on Opus 4.7 and later.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-opus-4-5-20251101", "name": "Claude Opus 4.5", "aliases": [ "claude-opus-4-5" ], "description": "The November 2025 Opus, and the oldest Opus still active. Its earliest retirement date of November 2026 is the nearest of any active Anthropic model — worth noting if it is still the default in your config.", "released_on": "2025-11-01", "earliest_shutdown_on": "2026-11-24", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-opus-4-6", "name": "Claude Opus 4.6", "aliases": [], "description": "The first Claude to use the dateless id format, where claude-opus-4-6 is itself a pinned snapshot rather than a pointer to a dated one. It is active, with an earliest retirement of February 2027, and still accepts the sampling parameters that later Opus releases reject.", "earliest_shutdown_on": "2027-02-05", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-opus-4-7", "name": "Claude Opus 4.7", "aliases": [], "description": "The release where Anthropic deprecated the sampling parameters themselves: temperature, top_p and top_k return a 400 error on this model and everything after it. That is a deprecation you hit at runtime rather than on a schedule, and it breaks more code than most model retirements do.", "earliest_shutdown_on": "2027-04-16", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-opus-4-8", "name": "Claude Opus 4.8", "aliases": [], "description": "The last Opus of the 4.x line, and the successor Anthropic names for Opus 4.1, Opus 4, Opus 3 and the Claude 2 family — four generations of migration paths converge here. Still active, with an earliest retirement date of May 2027.", "earliest_shutdown_on": "2027-05-28", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-opus-5", "name": "Claude Opus 5", "aliases": [], "description": "The recommended default for complex agentic coding and enterprise work, and the most common migration target on the Anthropic side of this catalog. Its earliest possible retirement is July 2027 — a longer guaranteed floor than any other model here.", "earliest_shutdown_on": "2027-07-24", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-sonnet-4-20250514", "name": "Claude Sonnet 4", "aliases": [ "claude-sonnet-4-0" ], "description": "The workhorse of the Claude 4 generation, and for a year the default in Claude Code. Retired June 15, 2026 in the same announcement as Opus 4.", "released_on": "2025-05-14", "deprecated_on": "2026-04-14", "shutdown_on": "2026-06-15", "status": "retired", "replacements": [ { "provider": "anthropic", "model": "claude-sonnet-4-6", "recommended": true, "note": "Sonnet 4.6 keeps the same pricing and raises the context window to 1M tokens. Its adaptive thinking replaces the explicit thinking budget Sonnet 4 used, so any code setting thinking.budget_tokens needs review.", "external": false } ], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "anthropic", "model": "claude-sonnet-4-5-20250929", "name": "Claude Sonnet 4.5", "aliases": [ "claude-sonnet-4-5" ], "description": "The September 2025 Sonnet, still active but with the second-nearest earliest retirement date of any live Anthropic model. It is the last Sonnet to use explicit extended thinking rather than the adaptive thinking introduced in 4.6.", "released_on": "2025-09-29", "earliest_shutdown_on": "2026-09-29", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-sonnet-4-6", "name": "Claude Sonnet 4.6", "aliases": [], "description": "The Sonnet that replaced both Sonnet 4 and Sonnet 3.7 when those were retired, which makes it the busiest migration target in the Sonnet line. Active, with an earliest retirement of February 2027.", "earliest_shutdown_on": "2027-02-17", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "anthropic", "model": "claude-sonnet-5", "name": "Claude Sonnet 5", "aliases": [], "description": "The balanced tier of the Claude 5 generation, and the one carrying introductory pricing through August 2026. Not deprecated, with an earliest retirement date of June 2027.", "earliest_shutdown_on": "2027-06-30", "status": "active", "replacements": [], "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "title": "Anthropic model deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-1.0-flash", "name": "Gemini 1.0 Flash", "aliases": [], "description": "The low-latency tier of the first Gemini generation, and the start of the Flash naming that Google still uses. Like 1.0 Pro it left the API without a published retirement date, which is why this page carries a status but no shutdown day.", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-2.5-flash", "recommended": true, "note": "Flash has stayed the name for the fast tier across four generations, so this is the natural target — but check the current Flash id, since 2.5 Flash is itself scheduled for retirement.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" }, { "url": "https://github.com/techdevsynergy/llm-model-deprecation", "title": "llm-model-deprecation registry (third-party)", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-1.0-pro", "name": "Gemini 1.0 Pro", "aliases": [], "description": "The first Gemini available through an API, launched in December 2023 as Googles answer to GPT-4. It shipped before Google settled on the model naming and lifecycle conventions used since, and no shutdown date was ever published for it — the registry that tracks it records the model as superseded with no retirement date on file.", "released_on": "2023-12-13", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-2.5-pro", "recommended": true, "note": "Nothing about a 1.0-era integration survives unchanged: the SDK, the request shape and the safety-setting enums have all been replaced since.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" }, { "url": "https://github.com/techdevsynergy/llm-model-deprecation", "title": "llm-model-deprecation registry (third-party)", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-1.5-flash", "name": "Gemini 1.5 Flash", "aliases": [ "gemini-1.5-flash-002" ], "description": "The cheap, fast half of the 1.5 generation, and the model that pushed long-context pricing down across the whole market. As with 1.5 Pro, Google has removed it from its published deprecation tables, so the shutdown date here is sourced from the community llm-model-deprecation registry.", "released_on": "2024-05-14", "shutdown_on": "2026-06-17", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-2.5-flash", "recommended": true, "note": "2.5 Flash is the direct successor, but note it carries its own shutdown date of October 16, 2026 — migrating here buys about a year, not a decade.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" }, { "url": "https://github.com/techdevsynergy/llm-model-deprecation", "title": "llm-model-deprecation registry (third-party)", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-1.5-pro", "name": "Gemini 1.5 Pro", "aliases": [ "gemini-1.5-pro-002" ], "description": "The model that made a million-token context window a mainstream expectation, and for a year the reason teams chose Gemini at all. Google no longer lists it on its deprecations page — rows are removed once a model is long gone — so the shutdown date here comes from the community llm-model-deprecation registry rather than a Google page, and should be treated as less firm than the rest of this site.", "released_on": "2024-02-15", "shutdown_on": "2026-06-17", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-2.5-pro", "recommended": true, "note": "The 2.5 generation thinks before answering by default, which changes both latency and token cost. Budget with thinkingConfig rather than assuming 1.5-era numbers carry over.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" }, { "url": "https://github.com/techdevsynergy/llm-model-deprecation", "title": "llm-model-deprecation registry (third-party)", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.0-flash", "name": "Gemini 2.0 Flash", "aliases": [ "gemini-2.0-flash-001" ], "description": "The model that made native tool use and a million-token context standard at Flash pricing, and the default Gemini for most of 2025. It was shut down on June 1, 2026 — the largest single removal of live traffic in Gemini's history, since so many integrations had never moved off it.", "released_on": "2025-02-05", "shutdown_on": "2026-06-01", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.6-flash", "recommended": true, "note": "Google skips the whole 2.5 and 3.x Flash lineage and names 3.6 Flash directly. Response formats are compatible; the thinking configuration is not, since 2.0 Flash had none.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.0-flash-lite", "name": "Gemini 2.0 Flash-Lite", "aliases": [ "gemini-2.0-flash-lite-001" ], "description": "The cheapest Gemini of the 2.0 generation, aimed at classification and extraction where per-call cost dominated. Shut down June 1, 2026 alongside gemini-2.0-flash.", "released_on": "2025-02-25", "shutdown_on": "2026-06-01", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-flash-lite", "recommended": true, "note": "The Lite tier persisted across generations, so this is a like-for-like swap — the closest thing to a drop-in in the Gemini catalog.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.0-flash-lite-preview", "name": "Gemini 2.0 Flash-Lite (preview)", "aliases": [ "gemini-2.0-flash-lite-preview-02-05" ], "description": "The preview of the 2.0 Lite tier, published on the same day as the 2.0 Flash launch. It outlived its own stable successor's preview cycle and was shut down in December 2025.", "released_on": "2025-02-05", "shutdown_on": "2025-12-09", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-2.5-flash-lite", "recommended": true, "note": "Google names 2.5 Flash-Lite here, which is itself scheduled for October 2026. gemini-3.1-flash-lite is the target that does not need migrating again.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-flash", "name": "Gemini 2.5 Flash", "aliases": [], "description": "Probably the highest-volume Gemini model ever deployed: fast, cheap, and the first Flash with a configurable thinking budget. Active, but Google has published October 16, 2026 as the earliest date it could be retired.", "released_on": "2025-06-17", "earliest_shutdown_on": "2026-10-16", "status": "active", "replacements": [ { "provider": "google", "model": "gemini-3.6-flash", "recommended": true, "note": "Google skips 3.1 and 3.5 and names 3.6 Flash directly. The thinking-budget configuration introduced in 2.5 Flash has changed shape since — re-read the thinking docs rather than porting the config.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-2.5-flash-lite", "name": "Gemini 2.5 Flash-Lite", "aliases": [], "description": "The cheapest model of the 2.5 generation, built for classification and extraction at volume where the thinking budget could be set to zero. Active, with an earliest shutdown of October 16, 2026 alongside the rest of the 2.5 line.", "released_on": "2025-07-22", "earliest_shutdown_on": "2026-10-16", "status": "active", "replacements": [ { "provider": "google", "model": "gemini-3.1-flash-lite", "recommended": true, "note": "Note that Google points Flash-Lite at the 3.1 generation while pointing Flash at 3.6 — the Lite tiers are on their own release cadence.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-2.5-flash-lite-preview-09-2025", "name": "Gemini 2.5 Flash-Lite (preview 09-2025)", "aliases": [], "description": "A September 2025 preview of the Flash-Lite tier, notable mainly for its id format: it used a four-digit year where every other Gemini preview uses two. Shut down March 2026.", "released_on": "2025-09-25", "shutdown_on": "2026-03-31", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-flash-lite", "recommended": true, "note": "Google routes this straight to the 3.1 Lite tier rather than to the stable 2.5 Flash-Lite, which is itself scheduled for October 2026.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-flash-preview-05-20", "name": "Gemini 2.5 Flash (preview 05-20)", "aliases": [], "description": "The May 2025 preview of 2.5 Flash, and the first Gemini to expose a configurable thinking budget. Shut down November 2025, six months after release.", "released_on": "2025-05-20", "shutdown_on": "2025-11-18", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.6-flash", "recommended": true, "note": "The thinking-budget parameter this preview introduced has changed shape twice since; check the current thinking docs rather than porting config.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-flash-preview-09-25", "name": "Gemini 2.5 Flash (preview 09-25)", "aliases": [], "description": "The September 2025 preview snapshot of 2.5 Flash, released to test changes ahead of the stable model. Shut down February 2026.", "released_on": "2025-09-25", "shutdown_on": "2026-02-17", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.6-flash", "recommended": true, "note": "Google points this snapshot past the stable 2.5 Flash id straight to 3.6 Flash, so there is no reason to migrate to 2.5 Flash as an interim step.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-pro", "name": "Gemini 2.5 Pro", "aliases": [], "description": "The model that put Google back at the front of the field in mid-2025, with thinking enabled by default and a million-token context window. It is still active and widely deployed, but Google has published October 16, 2026 as its earliest shutdown — which for a model this widely used is the single most important date on this site.", "released_on": "2025-06-17", "earliest_shutdown_on": "2026-10-16", "status": "active", "replacements": [ { "provider": "google", "model": "gemini-3.1-pro-preview", "recommended": true, "note": "The successor still carries a -preview suffix, so a migration off 2.5 Pro currently means moving onto a preview id. Factor that into the timing.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-2.5-pro-preview-03-25", "name": "Gemini 2.5 Pro (preview 03-25)", "aliases": [], "description": "The first 2.5 Pro preview, from March 2025, and the release where Gemini started thinking before answering by default. It was widely benchmarked under this id, which is why it still appears in evaluation code. Shut down December 2025.", "released_on": "2025-03-25", "shutdown_on": "2025-12-02", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-pro-preview", "recommended": true, "note": "If you are reproducing a benchmark rather than shipping, note that no hosted model reproduces this snapshot's behaviour.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-pro-preview-05-06", "name": "Gemini 2.5 Pro (preview 05-06)", "aliases": [], "description": "The second 2.5 Pro preview, from May 2025. Google shipped three preview snapshots in as many months before the stable release, each a separate id, and retired all three together in December 2025.", "released_on": "2025-05-06", "shutdown_on": "2025-12-02", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-pro-preview", "recommended": true, "note": "Two generations separate these models. Treat this as a re-evaluation rather than a swap.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-2.5-pro-preview-06-05", "name": "Gemini 2.5 Pro (preview 06-05)", "aliases": [], "description": "The last of three 2.5 Pro previews, released twelve days before the stable model. All three were shut down on the same day in December 2025.", "released_on": "2025-06-05", "shutdown_on": "2025-12-02", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-pro-preview", "recommended": true, "note": "Google points every 2.5 Pro preview at 3.1 Pro rather than at stable 2.5 Pro, which is itself scheduled for retirement in October 2026.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-3-flash-preview", "name": "Gemini 3 Flash (preview)", "aliases": [], "description": "The first Flash model of the Gemini 3 generation. Google lists a recommended replacement for it but no shutdown date, which puts it in an unusual middle state: superseded but not yet scheduled.", "released_on": "2025-12-17", "status": "active", "replacements": [ { "provider": "google", "model": "gemini-3.6-flash", "recommended": true, "note": "Google names gemini-3.6-flash as the replacement even though no shutdown date has been set. Moving now avoids a rushed migration when one is.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-3-pro-preview", "name": "Gemini 3 Pro (preview)", "aliases": [], "description": "The first Pro model of the Gemini 3 generation, and one of the shortest-lived entries in this catalog: released November 2025, shut down March 2026, less than four months later. Preview ids on the Gemini API carry real shutdown risk.", "released_on": "2025-11-18", "shutdown_on": "2026-03-09", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-pro-preview", "recommended": true, "note": "The successor is also a preview id. On the Gemini API, preview does not mean experimental so much as short-lived — plan to move again.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-3.1-flash-lite", "name": "Gemini 3.1 Flash-Lite", "aliases": [], "description": "The Flash-Lite of the 3.1 generation, and the model Google names as the replacement for the 2.0 and 2.5 Lite tiers. It is not deprecated, but Google has already published May 2027 as the earliest date it could be retired.", "released_on": "2026-05-07", "earliest_shutdown_on": "2027-05-07", "status": "active", "replacements": [ { "provider": "google", "model": "gemini-3.5-flash-lite", "recommended": true, "note": "Named as the eventual successor. No action is needed yet — this model is current and fully supported.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-3.1-flash-lite-preview", "name": "Gemini 3.1 Flash-Lite (preview)", "aliases": [], "description": "The preview that became gemini-3.1-flash-lite. It ran for under three months before being shut down in favour of the stable id — the standard Gemini pattern, and the reason preview ids should never reach production config.", "released_on": "2026-03-03", "shutdown_on": "2026-05-25", "status": "retired", "replacements": [ { "provider": "google", "model": "gemini-3.1-flash-lite", "recommended": true, "note": "Dropping the \"-preview\" suffix is the entire migration. Behaviour is unchanged.", "external": false } ], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "google", "model": "gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro (preview)", "aliases": [], "description": "The current Pro-tier model, and the migration target Google names for gemini-2.5-pro and every retired 2.5 Pro preview. It has carried the \"-preview\" suffix since February 2026 without a stable id appearing — long enough that treating it as production is now the norm rather than the exception.", "released_on": "2026-02-19", "status": "active", "replacements": [], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-3.5-flash", "name": "Gemini 3.5 Flash", "aliases": [], "description": "The Flash release between 3.1 and 3.6, still active with no shutdown date published. Gemini Flash models have historically run about sixteen months from release to shutdown, which is the useful planning number if you are pinning one.", "released_on": "2026-05-19", "status": "active", "replacements": [], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-3.5-flash-lite", "name": "Gemini 3.5 Flash-Lite", "aliases": [], "description": "The cheapest tier of the current Gemini generation, and the named successor to gemini-3.1-flash-lite. No shutdown date published yet.", "released_on": "2026-07-21", "status": "active", "replacements": [], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "google", "model": "gemini-3.6-flash", "name": "Gemini 3.6 Flash", "aliases": [], "description": "The current Flash model, and the successor Google names for the entire 2.x Flash line. Google publishes a shutdown date for almost every Gemini model as a matter of policy; this one has none yet, which is the clearest signal available that it is at the front of the queue rather than the back.", "released_on": "2026-07-21", "status": "active", "replacements": [], "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/deprecations", "title": "Gemini API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "ada", "name": "ada", "aliases": [], "description": "The smallest and fastest GPT-3 model, priced so low it was used for embeddings and simple classification at a scale nothing else made sense for. Retired January 2024; the embeddings side of its job moved to the dedicated text-embedding models.", "released_on": "2020-06-11", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "babbage-002", "recommended": true, "note": "For embeddings specifically, text-embedding-3-small is the modern equivalent and is not deprecated.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "babbage", "name": "babbage", "aliases": [], "description": "The second-smallest GPT-3 base model, mostly used for semantic search and cheap classification where latency mattered more than nuance. Retired January 2024 with the rest of the original GPT-3 line.", "released_on": "2020-06-11", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "babbage-002", "recommended": true, "note": "babbage-002 was the direct replacement and is deprecated in turn, shutting down September 2026.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "babbage-002", "name": "babbage-002", "aliases": [], "description": "A base completions model kept in service almost entirely as a fine-tuning target for classification work, where a small model trained on your own labels still beat prompting a large one. Its September 2026 shutdown ends that option.", "released_on": "2023-08-22", "deprecated_on": "2025-09-26", "shutdown_on": "2026-09-28", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.4-mini", "recommended": true, "note": "Existing fine-tunes do not transfer. Budget for re-running training on the new base and re-validating accuracy — a base-model fine-tune and an instruction-model fine-tune do not behave the same way.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "chatgpt-4o-latest", "name": "ChatGPT-4o latest", "aliases": [], "description": "The id that exposed the exact GPT-4o variant serving ChatGPT, complete with the personality tuning the API models did not have. It was popular for chat products that wanted ChatGPT's voice, and it was retired in February 2026.", "deprecated_on": "2025-11-18", "shutdown_on": "2026-02-17", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.1-chat-latest", "recommended": true, "note": "OpenAI named gpt-5.1-chat-latest as the successor — but that id has since been retired too. The chain currently ends at gpt-5.6-sol.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "chatgpt-image-latest", "name": "ChatGPT Image latest", "aliases": [], "description": "A floating pointer to whichever image model ChatGPT was using, for products that wanted to match what users saw in the app. Like every other -latest pointer OpenAI has shipped, it was retired on a shorter notice than the pinned ids it pointed at.", "deprecated_on": "2026-06-02", "shutdown_on": "2026-12-01", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "Move to a pinned id. Every floating pointer in this catalog — chat, codex and image alike — has now been deprecated at least once.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "code-davinci-002", "name": "code-davinci-002", "aliases": [], "description": "The Codex model behind the original GitHub Copilot, and the shortest notice OpenAI has ever given: deprecated on March 20, 2023 and shut down three days later. The backlash from researchers who lost access mid-study is a large part of why every provider now publishes deprecation schedules months ahead.", "deprecated_on": "2023-03-20", "shutdown_on": "2023-03-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-4o", "recommended": true, "note": "For code work today, gpt-5.3-codex is the direct descendant of this line. gpt-4o is what OpenAI named at the time.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "codex-mini-latest", "name": "codex-mini-latest", "aliases": [], "description": "The small coding model behind the first Codex CLI, and a floating pointer rather than a pinned snapshot. Retired February 2026, a day short of three months after the announcement.", "deprecated_on": "2025-11-17", "shutdown_on": "2026-02-12", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5-codex-mini", "recommended": true, "note": "gpt-5-codex-mini keeps the shell-tool integration this model was built around, so agent scaffolding usually transfers unchanged.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "computer-use-preview", "name": "computer-use-preview", "aliases": [ "computer-use-preview-2025-03-11" ], "description": "The model behind OpenAI's Operator agent, which drove a browser by emitting mouse and keyboard actions against screenshots. It stayed a preview for its entire life and was retired in July 2026 without a direct replacement.", "released_on": "2025-03-11", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "There is no drop-in computer-use id in the current generation. The computer_use tool schema does not carry over — this is a rewrite, not a swap.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "curie", "name": "curie", "aliases": [], "description": "The mid-size GPT-3 base model, roughly a tenth of davinci's price and the usual choice for classification once you had few-shot examples that worked. Its January 2024 shutdown removed the middle of the GPT-3 size ladder.", "released_on": "2020-06-11", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "davinci-002", "recommended": true, "note": "OpenAI collapsed four sizes into two. There is no curie-sized base model on the platform now.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "dall-e-2", "name": "DALL·E 2", "aliases": [], "description": "The model that put text-to-image generation in front of a general audience in 2022, complete with the waitlist that made it feel like an event. By the time it was retired in May 2026 it had been outclassed for years — it survived on inertia and on the inpainting API that had no direct successor for a while.", "released_on": "2022-04-06", "deprecated_on": "2025-11-14", "shutdown_on": "2026-05-12", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "The request shape differs: gpt-image models are called through the images endpoint with a different parameter set, and there is no direct equivalent of the DALL·E 2 variations call.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "dall-e-3", "name": "DALL·E 3", "aliases": [], "description": "The image model that finally rendered text inside images legibly and took prompts as prose rather than keyword soup. It rewrote your prompt before generating, which produced better pictures and made exact reproduction impossible — a trade the gpt-image line kept.", "released_on": "2023-10-03", "deprecated_on": "2025-11-14", "shutdown_on": "2026-05-12", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "Quality is a clear step up, but the automatic prompt rewriting behaves differently. Anything with prompts tuned against DALL·E 3s rewriter needs re-testing rather than porting.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "davinci", "name": "davinci", "aliases": [], "description": "The largest GPT-3 base model, and the one the 2020 paper actually described. It had no instruction tuning at all — you steered it with few-shot examples, which is the technique the whole prompt-engineering field grew out of. Retired January 2024.", "released_on": "2020-06-11", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "davinci-002", "recommended": true, "note": "davinci-002 was the named replacement base model and is itself deprecated, shutting down September 2026 — after which OpenAI hosts no base models at all.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "davinci-002", "name": "davinci-002", "aliases": [], "description": "The larger of the two remaining base completions models, and the last descendant of the original davinci line still callable. Deprecated with a September 2026 shutdown.", "released_on": "2023-08-22", "deprecated_on": "2025-09-26", "shutdown_on": "2026-09-28", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.4-mini", "recommended": true, "note": "Nothing on the current platform offers raw base-model continuation. If your use case depends on it rather than on instruction following, an open-weight model is a closer match than anything OpenAI still hosts.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-3.5-turbo", "name": "GPT-3.5 Turbo", "aliases": [ "gpt-3.5-turbo-0125", "gpt-3.5-turbo-completions" ], "description": "The model that made the chat completions API a commodity: cheap enough to put behind a free product, good enough that most people stopped writing prompt chains. It survived every deprecation wave from 2023 to 2026 because so much code was written against it — and it is finally going in October 2026.", "released_on": "2023-03-01", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "The biggest gotcha is max_tokens: current models use max_completion_tokens and reject the old parameter. Prompts tuned for GPT-3.5 are usually far more verbose than a 5.6-generation model needs.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-3.5-turbo-0301", "name": "GPT-3.5 Turbo (0301)", "aliases": [], "description": "The launch snapshot of GPT-3.5 Turbo, from March 2023 — the model that made chat APIs cheap and kicked off the current wave of applications. It was deprecated three months after release and took eighteen months to actually go.", "released_on": "2023-03-01", "deprecated_on": "2023-06-13", "shutdown_on": "2024-09-13", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo", "recommended": true, "note": "Now deprecated in turn. Anything still pinned to a 2023 snapshot should go straight to a current model rather than to the pointer.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-3.5-turbo-0613", "name": "GPT-3.5 Turbo (0613)", "aliases": [ "gpt-3.5-turbo-16k-0613" ], "description": "The June 2023 snapshot that introduced function calling to the API — the feature every agent framework since has been built on. Retired September 2024 along with its 16k variant.", "released_on": "2023-06-13", "deprecated_on": "2023-11-06", "shutdown_on": "2024-09-13", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo", "recommended": true, "note": "The undated pointer was the named successor and is itself now deprecated, shutting down October 2026. Plan for gpt-5.6-terra instead.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-3.5-turbo-1106", "name": "GPT-3.5 Turbo (1106)", "aliases": [], "description": "The November 2023 GPT-3.5 snapshot, the first to support JSON mode and parallel function calling. It is deprecated with a September 2026 shutdown, a month before the undated gpt-3.5-turbo pointer follows it.", "released_on": "2023-11-06", "deprecated_on": "2025-09-26", "shutdown_on": "2026-09-28", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.4-mini", "recommended": true, "note": "Swap max_tokens for max_completion_tokens, and replace JSON mode with structured outputs, which enforce your schema rather than only valid JSON.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "name": "GPT-3.5 Turbo Instruct", "aliases": [], "description": "The last completions-endpoint model OpenAI shipped, kept alive years past the rest of the /v1/completions family because logprobs and raw continuation behaviour had no equivalent in chat. Its September 2026 shutdown closes out the completions era entirely.", "released_on": "2023-09-18", "deprecated_on": "2025-09-26", "shutdown_on": "2026-09-28", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.4-mini", "recommended": true, "note": "This is an endpoint change, not just an id change: /v1/completions gives way to chat completions, prompts become messages, and raw continuation behaviour has no direct equivalent.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4", "name": "GPT-4", "aliases": [ "gpt-4-0613", "gpt-4-completions", "gpt-4-0613-completions" ], "description": "The model that started the current era, and still the id people type when they mean \"the good one\". OpenAI kept it alive for three and a half years — far longer than anything released since — and set its shutdown for October 2026.", "released_on": "2023-03-14", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol is a reasoning model, so it is not a behavioural drop-in: swap max_tokens for max_completion_tokens and expect reasoning tokens on the bill. Prompts written to coax GPT-4 into thinking step by step are now counterproductive.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4-0125-preview", "name": "GPT-4 Turbo preview (0125)", "aliases": [], "description": "The January 2024 preview snapshot, released to fix a \"laziness\" regression where the model truncated long code generations. Retired in March 2026.", "released_on": "2024-01-25", "deprecated_on": "2025-09-26", "shutdown_on": "2026-03-26", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Nothing about the preview snapshots carries forward; the response-truncation behaviour they existed to fix has not recurred.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4-0314", "name": "GPT-4 (0314)", "aliases": [], "description": "The launch snapshot of GPT-4, from March 2023. It was scheduled for removal twice — first for June 2024, then extended — before finally going in March 2026, which made it one of the longest-lived pinned snapshots OpenAI ever shipped.", "released_on": "2023-03-14", "deprecated_on": "2025-09-26", "shutdown_on": "2026-03-26", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-4", "recommended": true, "note": "The undated gpt-4 id is the smallest change, though it is itself deprecated with an October 2026 shutdown. For a migration that lasts, go to gpt-5.6-sol.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4-1106-preview", "name": "GPT-4 Turbo preview (1106)", "aliases": [], "description": "The November 2023 preview that introduced 128k context and JSON mode, and the one entry on this site with two conflicting shutdown dates: OpenAI first set it for March 26, 2026, then listed the same id again in the April 2026 announcement with a new date of October 23, 2026. The later announcement is the one that stands.", "released_on": "2023-11-06", "deprecated_on": "2025-09-26", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "JSON mode has been superseded by structured outputs, which enforce a schema rather than only guaranteeing valid JSON. Worth adopting during the move instead of after it.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4-32k", "name": "GPT-4 32k", "aliases": [ "gpt-4-32k-0613", "gpt-4-32k-0314" ], "description": "The 32,000-token variant of the original GPT-4, sold at twice the price of the 8k model when a long context was still a premium feature. Access was gated behind a waitlist for most of its life, so it never had the reach of plain gpt-4. GPT-4o made it redundant: 128k of context, faster, and cheaper than the 8k model it replaced.", "released_on": "2023-06-13", "deprecated_on": "2024-06-06", "shutdown_on": "2025-06-06", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-4o", "recommended": true, "note": "GPT-4o takes 128k of context, so nothing that fit in 32k needs chunking. It is a Chat Completions model with the same message format — the only required change is the id.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4-turbo", "name": "GPT-4 Turbo", "aliases": [ "gpt-4-turbo-2024-04-09", "gpt-4-turbo-completions" ], "description": "The 128k-context, cheaper GPT-4 that made long documents practical and held the default slot until GPT-4o arrived a month after its final snapshot. It is deprecated with an October 2026 shutdown.", "released_on": "2024-04-09", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Context is no longer the constraint it was, so chunking logic written for 128k can usually be deleted rather than ported.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4-turbo-preview", "name": "GPT-4 Turbo preview", "aliases": [ "gpt-4-turbo-preview-completions" ], "description": "A floating pointer that resolved to whichever GPT-4 Turbo preview snapshot was current, which meant its behaviour changed underneath anyone who used it. It was retired in March 2026 along with the snapshots it pointed at.", "deprecated_on": "2025-09-26", "shutdown_on": "2026-03-26", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Prefer a pinned id this time. Floating preview pointers have been retired in every OpenAI deprecation wave since 2024.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4-vision-preview", "name": "GPT-4 Vision preview", "aliases": [ "gpt-4-1106-vision-preview" ], "description": "The first GPT-4 that could see, and for a few months the only way to get image understanding from OpenAI. It was a separate id with its own rate limits and no function calling — constraints that disappeared when GPT-4o folded vision into the main model. Retired December 2024.", "released_on": "2023-11-06", "deprecated_on": "2024-06-06", "shutdown_on": "2024-12-06", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-4o", "recommended": true, "note": "Vision is no longer a separate id. The image_url content block carries over unchanged, and function calling now works alongside it.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4.1", "name": "GPT-4.1", "aliases": [], "description": "The 2025 refresh that fixed what GPT-4o was weakest at — instruction following and long-context recall — and became the default non-reasoning model for a year. Its nano sibling has been deprecated; gpt-4.1 and gpt-4.1-mini have not.", "released_on": "2025-04-14", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-4.1-mini", "name": "GPT-4.1 mini", "aliases": [], "description": "The mid-size GPT-4.1 variant, and one of the few 2025-era ids that has survived the 2026 deprecation waves untouched. It is not listed on the OpenAI deprecations page in any announcement, unlike gpt-4.1-nano.", "released_on": "2025-04-14", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-4.1-nano", "name": "GPT-4.1 nano", "aliases": [ "gpt-4.1-nano-2025-04-14" ], "description": "The cheapest model of the GPT-4.1 generation, and the only member of that family to be deprecated so far — gpt-4.1 and gpt-4.1-mini remain active. It was a common choice for classification and routing at high volume.", "released_on": "2025-04-14", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-luna", "recommended": true, "note": "Fine-tuned gpt-4.1-nano deployments are directed to gpt-5.4-nano instead; a fine-tune does not transfer and has to be re-run on the new base.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4.5-preview", "name": "GPT-4.5 preview", "aliases": [], "description": "The largest non-reasoning model OpenAI ever shipped, released in February 2025 and withdrawn from the API three months later — the shortest-lived flagship in the platform's history. It was expensive to serve and was overtaken by reasoning models that got more from less compute.", "released_on": "2025-02-27", "deprecated_on": "2025-04-14", "shutdown_on": "2025-07-14", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-4.1", "recommended": true, "note": "gpt-4.1 was the named successor and is still active — one of the few migration targets from this era that has not itself been deprecated.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o", "name": "GPT-4o", "aliases": [], "description": "The omni model that made multimodal input routine and halved the price of GPT-4 Turbo at the same time. It is still the default many production stacks reach for, and OpenAI has kept the floating gpt-4o pointer alive even as the dated snapshot behind it changed. The May 2024 snapshot is deprecated; the pointer itself is not.", "released_on": "2024-05-13", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-4o-2024-05-13", "name": "GPT-4o (2024-05-13)", "aliases": [], "description": "The original GPT-4o snapshot, the one that shipped at the May 2024 launch. Only this dated snapshot is deprecated — the floating gpt-4o pointer is not, and now resolves to a later snapshot. If you pinned the date, this is your deadline; if you did not, you have already been migrated.", "released_on": "2024-05-13", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "If the pin was defensive rather than deliberate, moving to the floating gpt-4o id is the smaller change and buys time.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-4o-audio-preview", "name": "GPT-4o Audio preview", "aliases": [], "description": "Audio in and audio out over ordinary Chat Completions, without the WebSocket session the realtime models require. It suited anything that did not need to be interrupted mid-sentence — voice notes, summaries, asynchronous replies.", "deprecated_on": "2025-09-15", "shutdown_on": "2026-05-07", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-audio-1.5", "recommended": true, "note": "Same request/response shape, so this is closer to a drop-in than the realtime migration. Re-check the supported input and output audio formats.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-audio-preview-2024-10-01", "name": "GPT-4o Audio preview (2024-10-01)", "aliases": [], "description": "The first audio-capable Chat Completions snapshot, shipped alongside the original realtime preview. It was retired in October 2025, a year to the day after release.", "released_on": "2024-10-01", "deprecated_on": "2025-06-10", "shutdown_on": "2025-10-10", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-audio-1.5", "recommended": true, "note": "The audio content-block format changed after this snapshot; check how you encode and read back audio parts.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-mini", "name": "GPT-4o mini", "aliases": [], "description": "The model that made GPT-4-class quality cheap enough to put in a loop, and for a long stretch the highest-volume id on the OpenAI platform. The floating pointer is not deprecated; only the dated gpt-4o-2024-05-13 snapshot of its larger sibling is.", "released_on": "2024-07-18", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-4o-mini-audio-preview", "name": "GPT-4o mini Audio preview", "aliases": [], "description": "The cost tier for audio on Chat Completions, aimed at transcription-adjacent work at volume. Retired in May 2026 with the rest of the gpt-4o audio family.", "deprecated_on": "2025-09-15", "shutdown_on": "2026-05-07", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-audio-mini", "recommended": true, "note": "gpt-audio-mini is deprecated in turn, shutting down January 2027. If you are migrating now, gpt-audio-1.5 avoids doing this twice.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-mini-realtime-preview", "name": "GPT-4o mini Realtime preview", "aliases": [], "description": "The cheaper realtime preview, which made always-on voice affordable enough for consumer products — realtime audio is billed per minute, so the tier below the flagship mattered more here than anywhere else.", "deprecated_on": "2025-09-15", "shutdown_on": "2026-05-07", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-realtime-mini", "recommended": true, "note": "Note that gpt-realtime-mini is itself deprecated with a January 2027 shutdown; gpt-realtime-2.1-mini is the target that lasts.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-mini-search-preview-2025-03-11", "name": "GPT-4o mini search-preview", "aliases": [], "description": "The cost tier of the search-preview pair, for grounded answers at volume. Retired on the same schedule as its larger sibling once web search became a tool rather than a model variant.", "released_on": "2025-03-11", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "Use the web_search tool on a current model. Search calls are billed separately from tokens, so re-model the unit cost before rolling out.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-realtime-preview", "name": "GPT-4o Realtime preview", "aliases": [ "gpt-4o-realtime-preview-2025-06-03", "gpt-4o-realtime-preview-2024-12-17" ], "description": "The model that introduced speech-to-speech over a WebSocket, with interruption handling and sub-second response — the first time a voice agent could be built without stitching together transcription, a chat model and text-to-speech. It stayed a preview for its entire eighteen-month life.", "released_on": "2024-10-01", "deprecated_on": "2025-09-15", "shutdown_on": "2026-05-07", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-realtime-1.5", "recommended": true, "note": "The session event schema changed between preview and GA. Expect to update event handlers, not just the model id.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-realtime-preview-2024-10-01", "name": "GPT-4o Realtime preview (2024-10-01)", "aliases": [], "description": "The original realtime snapshot from the October 2024 launch, deprecated eight months later on a four-month notice. Realtime previews were versioned by date and turned over faster than anything else on the platform at the time.", "released_on": "2024-10-01", "deprecated_on": "2025-06-10", "shutdown_on": "2025-10-10", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-realtime-1.5", "recommended": true, "note": "Two protocol revisions separate this snapshot from the current realtime model; treat the voice pipeline as a rewrite.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-4o-search-preview-2025-03-11", "name": "GPT-4o search-preview", "aliases": [], "description": "A GPT-4o variant with web search wired in at the model level, so a single call returned an answer with citations. OpenAI later moved search to a tool available on every model, which made the dedicated id redundant.", "released_on": "2025-03-11", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "Enable the web_search tool on any current model. The response shape differs: citations arrive as tool output rather than inline annotations.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5", "name": "GPT-5", "aliases": [ "gpt-5-2025-08-07" ], "description": "The August 2025 launch that unified the GPT and o-series lines into one model with a built-in router, and the id an enormous amount of production code was written against. OpenAI deprecated the snapshot behind it in June 2026 with a six-month runway. If your code says \"gpt-5\" with no date, this is the entry that applies to you.", "released_on": "2025-08-07", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol is a reasoning model, so budget output with max_completion_tokens rather than max_tokens, and expect reasoning_effort to dominate both cost and latency. Re-tune it before comparing prices.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5-chat-latest", "name": "GPT-5 chat-latest", "aliases": [], "description": "The first of the chat-latest pointers, tracking whatever non-reasoning model ChatGPT was serving. It was retired in July 2026 with three months notice, along with the entire first-generation Codex line.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol is pinned rather than floating, so behaviour stops changing underneath you between deploys.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5-codex", "name": "GPT-5 Codex", "aliases": [], "description": "The model that relaunched the Codex name for agentic coding, tuned for long autonomous runs rather than inline completion. It was retired in July 2026 together with every other first-generation Codex id in a single announcement.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "For coding-agent work the current equivalent is gpt-5.3-codex; OpenAI names Sol as the general replacement.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5-codex-mini", "name": "GPT-5 Codex mini", "aliases": [], "description": "The small coding model that replaced codex-mini-latest when that pointer was retired in February 2026. It kept the shell-tool integration the Codex line was built around, at a size that suits editor-loop completions rather than long-running agent work.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5-mini", "name": "GPT-5 mini", "aliases": [ "gpt-5-mini-2025-08-07" ], "description": "The cost tier of the original GPT-5 release, and the model that absorbed most of the volume GPT-4o mini used to carry. It was deprecated in the same June 2026 announcement as the rest of the GPT-5 family, on the same December shutdown date.", "released_on": "2025-08-07", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "Terra is the volume tier of the current generation, so this is the like-for-like swap. Check per-token pricing before assuming the migration is cost-neutral in either direction.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5-nano", "name": "GPT-5 nano", "aliases": [ "gpt-5-nano-2025-08-07" ], "description": "The smallest model of the GPT-5 generation, priced for classification and routing work that runs millions of times a day. Deprecated alongside gpt-5 and gpt-5-mini in June 2026.", "released_on": "2025-08-07", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-luna", "recommended": true, "note": "Luna occupies the same slot. At nano volumes even a small change in output length moves the bill, so re-measure token counts on real traffic rather than trusting the per-token price alone.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5-pro", "name": "GPT-5 Pro", "aliases": [ "gpt-5-pro-2025-10-06" ], "description": "The extended-compute tier of GPT-5, which spent far longer per request to buy accuracy on hard problems. It shipped two months after the base model and was deprecated on the same date, which is the useful lesson: a Pro tier does not get a longer life than the generation it belongs to.", "released_on": "2025-10-06", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol at a high reasoning_effort is the closest equivalent. There is no separate \"-pro\" id in the 5.6 generation — the compute dial moved into the request instead of the model name.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5.1-chat-latest", "name": "GPT-5.1 chat-latest", "aliases": [], "description": "The pointer that briefly replaced chatgpt-4o-latest when that id was retired in February 2026 — and was itself retired five months later. Anyone who followed OpenAI's own migration advice in late 2025 landed here and had to move again almost immediately.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Move to a pinned model rather than another chat-latest pointer; the pointer ids have now been retired three times in under a year.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5.1-codex", "name": "GPT-5.1 Codex", "aliases": [], "description": "The second-generation Codex model, and the one most coding agents shipped against through late 2025. Retired July 23, 2026 alongside gpt-5-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini and gpt-5.2-codex — four generations of the same line, gone on one day.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.3-codex", "recommended": true, "note": "gpt-5.3-codex is the current hosted Codex model. Given how fast this line turns over, pin it and set a calendar reminder rather than assuming a long life.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5.1-codex-max", "name": "GPT-5.1 Codex Max", "aliases": [], "description": "The largest Codex variant, built for repository-scale work where the model needed to hold an entire codebase in context. It lasted about eight months before being retired with the rest of the Codex line.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.3-codex", "recommended": true, "note": "There is no separate -max id in the 5.3 Codex generation; the capability moved into the base model.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5.1-codex-mini", "name": "GPT-5.1 Codex mini", "aliases": [], "description": "The small Codex model for editor-loop completions, where a 200ms difference in latency is more valuable than a smarter answer. Retired July 23, 2026.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "OpenAI names Terra rather than a Codex id here. For editor-latency work gpt-5.3-codex-spark is the closer fit.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5.2-chat-latest", "name": "GPT-5.2 chat-latest", "aliases": [], "description": "One of the non-reasoning chat pointers that mirrored the model behind ChatGPT at a given moment. These ids exist for conversational parity, not for tool use, and OpenAI retires them faster than the numbered API models — this one got a three-month notice, not six.", "deprecated_on": "2026-05-08", "shutdown_on": "2026-08-10", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "The chat-latest pointers do not support the full tool-calling surface, so moving to Sol is usually an upgrade rather than a like-for-like swap. Check any prompt that was tuned against ChatGPT-flavoured output.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5.2-codex", "name": "GPT-5.2 Codex", "aliases": [], "description": "The third-generation Codex model, and the shortest-lived of the family: it was deprecated within months of release and retired on the same day as its three predecessors.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.3-codex", "recommended": true, "note": "The direct successor. The Codex line has averaged well under a year per id — treat any Codex model as a moving target.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-5.3-chat-latest", "name": "GPT-5.3 chat-latest", "aliases": [], "description": "The successor pointer to gpt-5.2-chat-latest, deprecated on the same day and the same shutdown date as the id it replaced. Chat-latest pointers move with ChatGPT rather than with the API, which is exactly why pinning one is risky: the behaviour underneath can change without the id changing.", "deprecated_on": "2026-05-08", "shutdown_on": "2026-08-10", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol is a pinned model rather than a moving pointer, so output stays stable between deploys. That is the main behavioural difference to plan for.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-5.3-codex", "name": "GPT-5.3 Codex", "aliases": [], "description": "The current hosted Codex model, after gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max and gpt-5.2-codex were all retired together in July 2026. The Codex line turns over faster than any other family on the platform, which is the single most useful thing to know when pinning one of these ids.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.3-codex-spark", "name": "GPT-5.3 Codex Spark", "aliases": [], "description": "The low-latency Codex variant, built for the completion loop inside an editor rather than for long agent runs. It shares the Codex family's short release cadence: four of its predecessors were retired on a single day in July 2026.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.4", "name": "GPT-5.4", "aliases": [], "description": "For several months the default recommendation for professional work on the OpenAI platform, and still widely deployed. It is not deprecated, but two generations now sit above it, and OpenAI has been retiring GPT-5-era snapshots on roughly six-month notice.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.4-mini", "name": "GPT-5.4 mini", "aliases": [], "description": "The cost tier of the GPT-5.4 generation. It is the named replacement for the legacy fine-tuning base models — ft-gpt-3.5-turbo, ft-babbage-002 and ft-davinci-002 — which is an unusual role for a mini model and worth checking before assuming a fine-tune migrates cleanly.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.4-nano", "name": "GPT-5.4 nano", "aliases": [], "description": "The smallest model of the GPT-5.4 generation, and the successor OpenAI names for fine-tuned gpt-4.1-nano deployments. Nano-class models are where cost pressure is sharpest, which is also why they turn over faster than the tiers above them.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.4-pro", "name": "GPT-5.4 Pro", "aliases": [], "description": "The extended-compute variant of GPT-5.4, positioned for problems worth minutes rather than seconds. OpenAI retired the earlier gpt-5-pro snapshot on the same announcement as the rest of the GPT-5 family, so a Pro tier is no guarantee of a longer life than the base model.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.5", "name": "GPT-5.5", "aliases": [], "description": "The generation between GPT-5.4 and the Sol/Terra/Luna split, and the last one to use a plain numeric name. It has not been deprecated, but it sits directly behind models that have — worth knowing if a migration plan was written against it earlier in 2026.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.5-pro", "name": "GPT-5.5 Pro", "aliases": [], "description": "The extended-compute variant of GPT-5.5, which spends much longer per request in exchange for accuracy on problems where a wrong answer is expensive. Pro tiers have historically been deprecated on the same schedule as the base model they derive from, so its lifecycle is worth watching alongside GPT-5.5.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.6-luna", "name": "GPT-5.6 Luna", "aliases": [], "description": "The smallest and cheapest model of the GPT-5.6 generation, the slot GPT-4.1 nano used to occupy. OpenAI names it as the replacement for gpt-4.1-nano, so it inherits the classification, extraction and routing workloads that never needed a frontier model in the first place.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.6-sol", "name": "GPT-5.6 Sol", "aliases": [], "description": "The flagship of the Sol/Terra/Luna split, where OpenAI stopped shipping one frontier model in three sizes and started naming them for what they are for. Sol is the reasoning tier, and it is where almost every 2026 deprecation points: the o-series, the GPT-5 snapshots, the codex line and the chat-latest pointers all name it as their successor.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-5.6-terra", "name": "GPT-5.6 Terra", "aliases": [], "description": "The middle tier of the GPT-5.6 generation, aimed at high-volume work where latency and price matter more than the last few points of reasoning. It is the named replacement for o4-mini, the search-preview models and the long tail of GPT-3.5 ids, which makes it the busiest migration target on the OpenAI side of this catalog.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-audio", "name": "GPT Audio", "aliases": [], "description": "The GA audio model for Chat Completions, successor to the gpt-4o audio previews. Deprecated in July 2026 with a six-month runway, alongside the whole realtime and audio generation it belonged to.", "deprecated_on": "2026-07-20", "shutdown_on": "2027-01-20", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-audio-1.5", "recommended": true, "note": "A point upgrade within the same family — the closest thing to a drop-in in the audio line.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-audio-1.5", "name": "GPT Audio 1.5", "aliases": [], "description": "The current audio-in/audio-out model on Chat Completions, and the target OpenAI names for every deprecated audio id — the gpt-4o audio previews, gpt-audio and gpt-audio-mini all point here. Unlike the realtime models it is a normal request/response call.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-audio-mini", "name": "GPT Audio mini", "aliases": [], "description": "The cheap audio model, and the named replacement for gpt-4o-mini-audio-preview before being deprecated itself nine months later. Anyone following OpenAIs own migration advice through 2026 has now been moved twice.", "deprecated_on": "2026-07-20", "shutdown_on": "2027-01-20", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-audio-1.5", "recommended": true, "note": "There is no mini id in the 1.5 audio generation; the cheaper tier folded back into the base model.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-audio-mini-2025-10-06", "name": "GPT Audio mini (2025-10-06)", "aliases": [], "description": "The pinned October 2025 snapshot behind gpt-audio-mini, retired in July 2026 while the floating id it backed stayed alive until 2027. The dated snapshot going first is the opposite of what most people assume pinning buys them.", "released_on": "2025-10-06", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-audio-1.5", "recommended": true, "note": "Go straight to gpt-audio-1.5 rather than to the floating gpt-audio-mini, which shuts down in January 2027.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "gpt-image-1", "name": "GPT Image 1", "aliases": [], "description": "The first image model of the GPT era, and the one that made image editing with a reference picture routine. It replaced the DALL·E line and was itself deprecated less than a year later, in the same April 2026 announcement that took out the o-series.", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "A same-family upgrade, so the request shape carries over. Check output pricing per image size before rolling out — the tiers changed.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-image-1-mini", "name": "GPT Image 1 mini", "aliases": [], "description": "The cost tier of the first GPT image generation, for thumbnails and drafts where a cheaper picture is the right answer. Deprecated in June 2026 with a December shutdown, two months after its larger sibling goes.", "deprecated_on": "2026-06-02", "shutdown_on": "2026-12-01", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "There is no mini id in the gpt-image-2 generation. Cost control moves to the size and quality parameters instead of a separate model.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-image-1.5", "name": "GPT Image 1.5", "aliases": [], "description": "A late refresh of the first GPT image generation, deprecated barely after it landed — it and gpt-image-2 were announced close enough together that many teams never migrated onto it at all.", "deprecated_on": "2026-06-02", "shutdown_on": "2026-12-01", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-image-2", "recommended": true, "note": "If you are still on gpt-image-1, skip 1.5 entirely and go straight to gpt-image-2 — 1.5 shuts down first.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-image-2", "name": "GPT Image 2", "aliases": [], "description": "The current image model, and the migration target for every image id OpenAI has deprecated — DALL·E 2 and 3, gpt-image-1, its mini and 1.5 variants, and the chatgpt-image-latest pointer all converge here. The whole image line turned over inside eighteen months.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-oss-120b", "name": "gpt-oss-120b", "aliases": [], "description": "OpenAI's large open-weight model, released under Apache 2.0 and runnable on hardware you own. Open weights change the deprecation question entirely: OpenAI can stop hosting it, but nobody can shut down a copy you have already downloaded.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-oss-20b", "name": "gpt-oss-20b", "aliases": [], "description": "The smaller of the two gpt-oss open-weight releases, sized to run on a single consumer GPU. Like its larger sibling it carries no shutdown risk in the usual sense — the weights are Apache 2.0 licensed and already distributed.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-realtime", "name": "GPT Realtime", "aliases": [], "description": "The first realtime model to leave preview, in late 2025 — and deprecated less than a year later. Voice models have the shortest lifecycles on the platform, which matters more than usual because a realtime integration is a protocol, not just an id.", "deprecated_on": "2026-07-20", "shutdown_on": "2027-01-20", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-realtime-2.1", "recommended": true, "note": "Six months of notice, which is the most generous window in the current OpenAI deprecation wave. Use it — realtime migrations touch session handling, not just a model string.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-realtime-1.5", "name": "GPT Realtime 1.5", "aliases": [], "description": "The speech-to-speech model that replaced the gpt-4o realtime previews when those were shut down in May 2026. Realtime models use a WebSocket session rather than a request/response call, so migrating between them is a protocol change as much as an id change.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-realtime-2.1", "name": "GPT Realtime 2.1", "aliases": [], "description": "The current realtime model, named as the successor when OpenAI deprecated the entire gpt-realtime and gpt-audio generation in July 2026. It is the third realtime id in under two years.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-realtime-2.1-mini", "name": "GPT Realtime 2.1 mini", "aliases": [], "description": "The cost tier of the current realtime generation, for voice applications where per-minute audio pricing dominates the bill. It replaces gpt-realtime-mini, which is deprecated with a January 2027 shutdown.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "gpt-realtime-mini", "name": "GPT Realtime mini", "aliases": [], "description": "The cost tier of the first GA realtime generation, and the successor to gpt-4o-mini-realtime-preview. It was deprecated in July 2026, which puts anyone who migrated off the preview on their second voice migration in a year.", "deprecated_on": "2026-07-20", "shutdown_on": "2027-01-20", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-realtime-2.1-mini", "recommended": true, "note": "A same-family upgrade, so session events should carry over more cleanly than the preview-to-GA move did.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "gpt-realtime-mini-2025-10-06", "name": "GPT Realtime mini (2025-10-06)", "aliases": [], "description": "The pinned October 2025 snapshot of the realtime mini model. It was retired in July 2026, six months ahead of the floating gpt-realtime-mini id it sat behind — pinning a snapshot bought less time here, not more.", "released_on": "2025-10-06", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-realtime-2.1-mini", "recommended": true, "note": "Skip the floating gpt-realtime-mini id, which is itself deprecated with a January 2027 shutdown.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "o1", "name": "o1", "aliases": [ "o1-2024-12-17" ], "description": "The first generally available reasoning model, and the release that proved spending inference compute could substitute for a larger model. It introduced the constraints the whole o-series inherited: no system messages at launch, no streaming, and max_completion_tokens in place of max_tokens.", "released_on": "2024-12-17", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol lifts most of o1's restrictions — system messages, streaming and tool calls all work — so code written defensively around o1 can usually be simplified rather than ported.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o1-mini", "name": "o1-mini", "aliases": [], "description": "The cheap reasoning model of the first o-series generation, strong on maths and code for its size but noticeably weaker on general knowledge. Retired October 2025.", "released_on": "2024-09-12", "deprecated_on": "2025-04-28", "shutdown_on": "2025-10-27", "status": "retired", "replacements": [ { "provider": "openai", "model": "o4-mini", "recommended": true, "note": "o4-mini was the named successor and is now itself deprecated, shutting down October 2026. Two migrations in eighteen months for anyone who followed the advice each time.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "o1-preview", "name": "o1-preview", "aliases": [], "description": "The first public reasoning model, released as a preview in September 2024 and retired ten months later — the shortest life of any widely adopted OpenAI model. It hid its reasoning tokens but billed for them, which is where the industry-wide argument about reasoning-token pricing started.", "released_on": "2024-09-12", "deprecated_on": "2025-04-28", "shutdown_on": "2025-07-28", "status": "retired", "replacements": [ { "provider": "openai", "model": "o3", "recommended": true, "note": "OpenAI named o3 as the successor; o3 is itself now deprecated with a December 2026 shutdown, so the durable target is gpt-5.6-sol.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "o1-pro", "name": "o1-pro", "aliases": [ "o1-pro-2025-03-19" ], "description": "The long-compute variant of o1, available only through the Responses API and priced for problems where minutes of thinking are worth it. Deprecated in April 2026 with an October shutdown.", "released_on": "2025-03-19", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "OpenAI directs o1-pro traffic to Sol with reasoning set to its pro mode rather than to a separate model id. The dial moved from the id into the request body.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o3", "name": "o3", "aliases": [ "o3-2025-04-16" ], "description": "The reasoning model that made chain-of-thought a product feature rather than a prompting trick, and the last o-series flagship before the line was folded into GPT-5. It outlived o1 by more than a year and is still widely referenced in evaluation code and papers.", "released_on": "2025-04-16", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Both are reasoning models, so max_completion_tokens and reasoning_effort carry over. Reasoning tokens are billed as output on both, which is the line item to compare.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o3-deep-research", "name": "o3-deep-research", "aliases": [ "o3-deep-research-2025-06-26" ], "description": "A research-agent model that ran multi-step web investigations autonomously and returned a cited report. It was only ever available on the Responses API with the web-search tool enabled, and was retired in July 2026 with its o4-mini sibling.", "released_on": "2025-06-26", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Sol plus the web-search tool covers the same ground, but you now orchestrate the research loop yourself rather than getting it built into the model.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "o3-mini", "name": "o3-mini", "aliases": [ "o3-mini-2025-01-31" ], "description": "The reasoning model that was cheap enough to use by default, and for much of 2025 the best value on the platform for anything involving maths or code. It is deprecated with an October 2026 shutdown.", "released_on": "2025-01-31", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "reasoning_effort carries over directly. If o3-mini was chosen for price rather than capability, compare Terra as well before defaulting to Sol.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o3-pro", "name": "o3-pro", "aliases": [ "o3-pro-2025-06-10" ], "description": "The long-compute variant of o3, aimed at problems where an answer is worth minutes of thinking. It only ever ran on the Responses API, so migrations off it are usually simpler than they look — the calling code is already shaped for the successor.", "released_on": "2025-06-10", "deprecated_on": "2026-06-11", "shutdown_on": "2026-12-11", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "Raise reasoning_effort on Sol to recover o3-pro-like depth. Expect different latency characteristics: Sol returns sooner at equivalent effort than o3-pro did.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o4-mini", "name": "o4-mini", "aliases": [ "o4-mini-2025-04-16" ], "description": "The last of the mini reasoning models, and the named successor when o1-mini was retired in 2025 — which makes it the second stop on a migration path that now needs a third. Deprecated April 2026, shutting down October 2026.", "released_on": "2025-04-16", "deprecated_on": "2026-04-22", "shutdown_on": "2026-10-23", "status": "deprecated", "replacements": [ { "provider": "openai", "model": "gpt-5.6-terra", "recommended": true, "note": "Note that OpenAI points o4-mini at Terra, not at Sol: the replacement for a small reasoning model is now a general model, not a smaller reasoner.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "deprecated" }, { "provider": "openai", "model": "o4-mini-deep-research", "name": "o4-mini-deep-research", "aliases": [ "o4-mini-deep-research-2025-06-26" ], "description": "The cheaper deep-research model, sized for research loops run at volume rather than one report at a time. Retired July 23, 2026 alongside o3-deep-research.", "released_on": "2025-06-26", "deprecated_on": "2026-04-22", "shutdown_on": "2026-07-23", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-5.6-sol", "recommended": true, "note": "No deep-research id exists in the current generation. Expect to rebuild the search-and-summarise loop in application code.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "omni-moderation", "name": "omni-moderation", "aliases": [], "description": "The multimodal moderation endpoint that replaced the text-moderation family in October 2025. It scores images as well as text, and it is free — which is why the older text-only ids had no reason to survive.", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "text-ada-001", "name": "text-ada-001", "aliases": [], "description": "The smallest InstructGPT model, fast and nearly free, used for classification and simple extraction where a wrong answer was cheap. It was the bottom rung of the ladder that ran ada, babbage, curie, davinci — a naming scheme retired along with the models themselves.", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "The named successor kept the completions endpoint alive, but it is deprecated too, shutting down September 28, 2026. After that no completions-style model remains on the platform.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-babbage-001", "name": "text-babbage-001", "aliases": [], "description": "The second rung of the InstructGPT ladder, a step up from ada for classification work that needed slightly more nuance. Retired in January 2024 when OpenAI collapsed four base sizes into two.", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "The named successor kept the completions endpoint alive, but it is deprecated too, shutting down September 28, 2026. After that no completions-style model remains on the platform.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-curie-001", "name": "text-curie-001", "aliases": [], "description": "The mid-tier InstructGPT model, and for a while the sweet spot on price and quality for summarisation before GPT-3.5 Turbo undercut the whole ladder at once. Retired January 2024.", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "The named successor kept the completions endpoint alive, but it is deprecated too, shutting down September 28, 2026. After that no completions-style model remains on the platform.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-davinci-001", "name": "text-davinci-001", "aliases": [], "description": "The first instruction-tuned davinci, and the model that demonstrated that following an instruction beat completing a pattern. It was quickly overtaken by text-davinci-002 and -003, but the id survived in tutorials for years after it stopped being the right choice.", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "The named successor kept the completions endpoint alive, but it is deprecated too, shutting down September 28, 2026. After that no completions-style model remains on the platform.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-davinci-002", "name": "text-davinci-002", "aliases": [], "description": "The first InstructGPT model trained with supervised fine-tuning on human demonstrations, and the immediate predecessor to text-davinci-003. It is the model most 2022-era research papers benchmarked against, which is why the id still turns up in code long after its January 2024 retirement.", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "No current model reproduces this one's behaviour. If you are reproducing a paper rather than shipping a product, an open-weight base model is a closer comparison.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-davinci-003", "name": "text-davinci-003", "aliases": [], "description": "The InstructGPT model that powered the first ChatGPT preview and most of the original prompt-engineering literature. It was a completions model — you sent a prompt string and got a continuation — and its retirement in January 2024 is what forced a generation of code onto the chat API.", "released_on": "2022-11-28", "deprecated_on": "2023-07-06", "shutdown_on": "2024-01-04", "status": "retired", "replacements": [ { "provider": "openai", "model": "gpt-3.5-turbo-instruct", "recommended": true, "note": "The named successor kept the completions endpoint alive; it is now deprecated too, shutting down September 2026. After that, no completions-style model remains.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" }, { "provider": "openai", "model": "text-embedding-3-large", "name": "text-embedding-3-large", "aliases": [], "description": "The large embedding model, and the strongest retrieval quality OpenAI offers. Like its smaller sibling it is absent from the deprecations page: an embedding model cannot be retired on a chat-model schedule, because doing so would invalidate every vector index built against it.", "released_on": "2024-01-25", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "text-embedding-3-small", "name": "text-embedding-3-small", "aliases": [], "description": "The small embedding model, cheaper and stronger than the ada-002 it succeeded, with configurable output dimensions. Embedding models age differently from chat models: changing one invalidates every vector already in your index, so OpenAI has kept this line stable far longer than the chat line.", "released_on": "2024-01-25", "status": "active", "replacements": [], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "text-embedding-ada-002", "name": "text-embedding-ada-002", "aliases": [], "description": "The embedding model behind most retrieval systems built between 2022 and 2024, and one of the most commonly assumed-dead ids on the platform. It is not on OpenAI's deprecations page: re-embedding a corpus is expensive enough that the old vectors have to keep working, so this one has outlived every chat model of its era.", "released_on": "2022-12-15", "status": "active", "replacements": [ { "provider": "openai", "model": "text-embedding-3-small", "recommended": false, "note": "Not a deprecation — but 3-small is cheaper and scores better. Only worth it if you can afford to re-embed everything, since vectors from the two models are not comparable.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "active" }, { "provider": "openai", "model": "text-moderation-007", "name": "text-moderation-007", "aliases": [ "text-moderation-stable", "text-moderation-latest" ], "description": "The last of the text-only moderation classifiers, and the endpoint most trust-and-safety pipelines were built against before images became table stakes. Both floating pointers, -stable and -latest, resolved to it and went down with it in October 2025.", "deprecated_on": "2025-04-28", "shutdown_on": "2025-10-27", "status": "retired", "replacements": [ { "provider": "openai", "model": "omni-moderation", "recommended": true, "note": "The category set is wider and the response includes per-category scores for images as well as text, so any thresholds tuned against the old categories need re-checking rather than copying.", "external": false } ], "sources": [ { "url": "https://developers.openai.com/api/docs/deprecations", "title": "OpenAI API deprecations", "accessed": "2026-07-30" } ], "last_verified": "2026-07-30", "computed_status": "retired" } ] }