OpenAI model deprecation is the process by which OpenAI retires models and endpoints from its API. OpenAI sells no maintenance product, so for API customers deprecation plays the role that end of support plays for on-premises software. OpenAI’s deprecations page states: “As we launch safer and more capable models, we regularly retire older models. Software relying on OpenAI models may need occasional updates to keep working.”[1]
For an asset manager, the practical task is to know which applications call which model snapshot, and when each snapshot will shut down. Contract terms on updates and beta features set what the customer can do when a change affects it.[2]
Lifecycle states
OpenAI uses three terms (rule Deprecated models stop being accessible at the shutdown date; legacy models no longer receive updates).[1]
| State | Meaning |
|---|---|
| Legacy | Models and endpoints “that no longer receive updates”. OpenAI says “You can expect that a legacy model or endpoint will be deprecated at some point in the future.” |
| Deprecated | A model or endpoint becomes deprecated as soon as the deprecation is announced. “All deprecated models and endpoints will also have a shut down date.” |
| Shut down (sunset) | The model or endpoint is no longer accessible. OpenAI uses “sunset” and “shut down” interchangeably. |
Source: OpenAI deprecations page, as retrieved on 2026-10-02.[1]
Notice periods
“Unless safety or compliance concerns require a faster timeline”, OpenAI gives these minimum notice periods before retiring a model (rule Model retirement notice: 6 months GA, 3 months specialized variants, shorter for previews):[1]
| Model type | Minimum notice | Examples given by OpenAI |
|---|---|---|
| Generally available models | At least 6 months | |
| Specialized variants of generally available models | At least 3 months | Chat variants such as gpt-5.1-chat-latest, Codex variants such as gpt-5.3-codex, deep research variants such as o3-deep-research |
| Preview models (named with “preview”) | Much shorter, such as 2 weeks | computer-use-preview, gpt-4o-audio-preview |
When OpenAI announces a deprecation, it notifies “customers who are actively using the model by email” and lists it on the deprecations page.[1] An organization that relies on email should make sure the notice reaches the team that owns the application, not only the account owner. If safety or compliance concerns require a faster retirement, OpenAI “will provide as much notice as reasonably possible”.[1]
OpenAI advises against preview models in production: “We don’t recommend using preview models for business-critical production workloads unless you can migrate on short notice.”[1] “In some cases, developers may be able to provision dedicated capacity for continued access after a model’s shutdown date”, through OpenAI sales.[1]
Recent and upcoming shutdowns
The deprecations page lists each announcement with its shutdown date and recommended replacement. Examples as of 2026-10-02:[1]
| Announced | What | Shutdown | Recommended replacement |
|---|---|---|---|
| 2026-10-01 | gpt-5.1, gpt-5.3-codex, gpt-5.4-nano | 2027-04-01 | gpt-6-sol; gpt-6-luna for gpt-5.4-nano |
| 2026-10-01 | Text-to-speech models tts-1, tts-1-hd and gpt-4o-mini-tts snapshots | 2027-01-06 | gpt-realtime-2.1-mini |
| 2026-08-26 | whisper-1 and gpt-4o transcription models | 2027-02-26 | gpt-live-transcribe or gpt-transcribe |
| 2026-07-20 | Legacy audio, realtime and transcription families | 2027-01-20 | gpt-realtime-2.1, gpt-audio-1.5 and others |
| 2026-06-11 | Older GPT-5 and o3 snapshots, such as gpt-5-2025-08-07 | 2026-12-11 | gpt-5.6-sol and others |
| 2026-06-03 | Reusable prompts (v1/prompts), Evals platform, Agent Builder | 2026-11-30 | Application code; Agents SDK or ChatGPT Workspace Agents |
| 2026-06-02 | Older GPT Image models, such as gpt-image-1-mini | 2026-12-01 | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
Deprecations cover platform features as well as models. For the Evals platform, existing evals become read-only on 2026-10-31 before the shutdown on 2026-11-30.[1] A model inventory should therefore also record which OpenAI platform features an application depends on.
Fine-tuned models
On 2026-05-07 OpenAI announced changes to its self-serve fine-tuning platform. Fine-tuning jobs were closed to organizations that had not run fine-tuning before. From 2026-07-02 they were closed to organizations with no inference on a fine-tuned model in the past 60 days. From 2027-01-06, “Active existing customers will no longer be able to create new fine-tuning jobs” (rule Self-serve fine-tuning closed to new jobs from 2027-01-06; inference continues until base model deprecation).[1]
Existing fine-tuned models keep running for a time: “Inference on fine-tuned models will continue to be available until the base models are deprecated.”[1] A fine-tuned model is therefore tied to its base model’s shutdown date. Past announcements also remove fine-tuned versions directly, for example ft-gpt-3.5-turbo and ft-gpt-4 on 2026-10-23.[1] Using OpenAI’s own fine-tuning service is a Permitted Exception to the restriction on using Output to build competing models.[2]
Contract terms
The Services Agreement does not refer to the deprecations page by name. Its general terms on updates and beta services apply.[2]
- Updates. “OpenAI may update the Services periodically.” “If an OpenAI update materially reduces the Services functionality, OpenAI will notify Customer at the Account email address.” The customer may then terminate the Agreement on thirty days’ written notice, but only “Within five business days of receipt of this notice”. The right “will not apply to updates made to features provided on a beta or evaluation basis” (rule Update that materially reduces functionality: termination right within five business days).[2]
- Warranty. OpenAI warrants that the Services “will conform in all material respects with the Documentation”. Documentation includes material that OpenAI “otherwise makes publicly available”.[2]
- Beta Services. Beta Services are features “identified as alpha, beta, preview, early access, or evaluation”. They “may not be supported and may be changed at any time without notice”, and OpenAI has no liability for them (rule Beta Services, including preview features, carry no support commitment or liability).[2] Under this definition, preview models are Beta Services. That matches the short notice period on the deprecations page.
The five-business-day window is short. A customer that wants to keep the termination right should route OpenAI notices to someone who can decide within that time. Price and plan terms are covered in OpenAI ChatGPT and API licensing, and the rest of the agreement in OpenAI Services Agreement and data controls.
Out of scope
- Model retirements inside ChatGPT, which follow product announcements rather than the API deprecations page.
- Scale Tier and dedicated capacity terms for keeping a model after its shutdown date, which are agreed with OpenAI sales.[1]
- The gpt-oss open-weight models, which the licensee runs itself and which do not shut down. See OpenAI licensing.