Deprecations

Which models are retiring soon

Sunset dates and recommended replacements across the 9 providers we track — researched directly from each provider's own lifecycle docs, with the providers that publish nothing usable flagged honestly rather than left out silently.

As of Aug 26, 2026. Hand-researched, not part of the automated daily pipeline — confirm against the source link before a production migration.

Upcoming retirements, soonest first

ProviderModel(s)RetiresRecommended replacement
OpenAIOpenAI
Assistants API
Aug 26, 2026
imminent
Responses API and Conversations API
GoogleGoogle
gemini-robotics-er-1.6-preview
Aug 31, 2026
~5d
gemini-robotics-er-2-preview
MistralMistral
Mistral Medium 3.1 (25.08)
Aug 31, 2026
~5d
Mistral Medium 3.5
MistralMistral
Mistral Medium 3 (25.05)
Aug 31, 2026
~5d
Mistral Medium 3.5
MoonshotMoonshot
Kimi K2.5 and Moonshot v1 series *
Aug 31, 2026
~5d
Kimi K3
OpenAIOpenAI
sora-2, sora-2-pro (all dated snapshots)
Sep 24, 2026
~29d
No replacement announced
OpenAIOpenAI
gpt-3.5-turbo-1106, gpt-3.5-turbo-instruct, babbage-002, davinci-002
Sep 28, 2026
~33d
gpt-5.6-terra
GoogleGoogle
gemini-2.5-flash-image
Oct 2, 2026
~37d
gemini-3.1-flash-image-preview
QwenQwen
Next scheduled retirement batch
Oct 10, 2026
~45d
Check DashScope console for affected models
OpenAIOpenAI
gpt-3.5-turbo-0125, gpt-4-0613, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o1-2024-12-17, o3-mini-2025-01-31
Oct 23, 2026
~58d
gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna (varies by model)
OpenAIOpenAI
gpt-5-2025-08-07, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07, gpt-5-pro-2025-10-06, o3-2025-04-16, o3-pro-2025-06-10
Dec 11, 2026
~107d
gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna (varies by model)
OpenAIOpenAI
gpt-realtime, gpt-audio, gpt-4o-realtime, gpt-4o-audio (and mini variants)
Jan 20, 2027
~147d
gpt-realtime-2.1 / gpt-audio-1.5 (varies by model)
GoogleGoogle
gemini-3.1-flash-lite
May 7, 2027
~254d
gemini-3.5-flash-lite
GoogleGoogle
gemini-embedding-001
May 14, 2028
~627d
gemini-embedding-2

Coverage by provider

OpenAIOpenAI
Publishes a lifecycle page
Official source ↗
AnthropicAnthropic
Publishes a lifecycle page

Anthropic commits to at least 60 days' notice before retiring a publicly released model.

Official source ↗
GoogleGoogle
Publishes a lifecycle page
Official source ↗
MistralMistral
Publishes a lifecycle page
Official source ↗
MoonshotMoonshot
Partial info only
Official source ↗
QwenQwen
Partial info only

Alibaba publishes retirement batch dates but not which specific models are in each batch — check the Model Observation page in your own DashScope console for whether your models are affected.

Official source ↗
xAI
No public page

No persistent deprecation index — xAI only publishes one-off migration pages when a specific model is retired, with no recurring page to check.

Official source ↗
DeepSeekDeepSeek
No public page

No dedicated deprecation page — retirement announcements, if any, appear only in the general changelog.

Official source ↗
Z.AI
No public page

No public deprecation page found anywhere in official docs.

Official source ↗

Why this list is shorter than you'd expect

Several providers' own deprecation pages list dozens of entries, but most of them are already past their retirement date by the time we check — we filter strictly to shutdown dates after this page's own "as of" date, so a model that already retired isn't listed here as if it were still upcoming. That's also why Anthropic currently shows zero upcoming entries: nothing on their published lifecycle table is between "deprecated" and "retired" right now, and they commit to at least 60 days' notice before changing that.

Why three providers show nothing at all

xAI, DeepSeek, and Z.AI don't publish a persistent, dedicated deprecation page the way the other six do — xAI only posts one-off migration pages when a specific retirement happens, DeepSeek folds any announcements into its general changelog, and Z.AI has no public deprecation documentation we could find anywhere. That's a real gap in what's publicly trackable, not an editorial choice on our part — see the rate-limits page for the same pattern of honestly disclosed gaps across providers.

Why this isn't a live daily feed

Unlike pricing, no provider (and no third-party aggregator we could find) publishes deprecation data in a machine-readable feed — every entry above was read directly off a provider's own docs page by hand. This page is refreshed periodically rather than daily; see methodology for which parts of the site are live-pipelined and which are dated snapshots like this one.

Frequently asked questions

Why doesn't this page show every deprecation a provider has ever announced?
Because most of them have already passed. We filter to only show entries whose retirement date is still in the future as of this page's own "as of" date — a model that already retired isn't useful information for planning a migration.
Why does Anthropic show zero upcoming retirements?
Because nothing on their published model lifecycle table is currently in the deprecated-but-not-yet-retired window. Anthropic commits to at least 60 days' notice before retiring a model, so this list will update as soon as they announce one.
What should I do if my provider isn't documented here?
For xAI, DeepSeek, and Z.AI, there's no public lifecycle page to check — your best options are the provider's own changelog/blog, their developer Discord or forum, or simply monitoring your own API responses for deprecation warnings, since these providers may announce a retirement with little public trace beforehand.