Model deprecations tracker
Models that API providers have retired or scheduled for shutdown, with the provider's recommended replacement and — where it exists — a place the same model is still served. Every row is checked against the provider's own deprecations page; last verified 2026-09-16.
| Shutdown | Provider | Model id | Scope | Provider's replacement | Same model still served |
|---|---|---|---|---|---|
| 2026-08-16 announced 2026-06-17 |
Groq | llama-3.3-70b-versatile |
Free and Developer tiers | openai/gpt-oss-120b or qwen/qwen3.6-27b | llama-3.3-70b-versatile (alias accepted) |
| 2026-08-16 announced 2026-06-17 |
Groq | llama-3.1-8b-instant |
Free and Developer tiers | openai/gpt-oss-20b | — |
| 2026-07-17 announced 2026-06-17 |
Groq | qwen/qwen3-32b |
Free and Developer tiers | openai/gpt-oss-120b | — |
| 2026-07-17 announced 2026-06-17 |
Groq | meta-llama/llama-4-scout-17b-16e-instruct |
Free and Developer tiers | openai/gpt-oss-120b or qwen/qwen3.6-27b | — |
| 2026-04-15 announced 2026-03-23 |
Groq | moonshotai/kimi-k2-instruct-0905 |
All self-serve tiers | openai/gpt-oss-120b | — |
| 2026-03-09 announced 2026-02-20 |
Groq | meta-llama/llama-4-maverick-17b-128e-instruct |
All self-serve tiers | openai/gpt-oss-120b | — |
| 2026-03-05 announced 2026-02-10 |
Groq | meta-llama/llama-guard-4-12b |
All self-serve tiers | openai/gpt-oss-safeguard-20b | — |
| 2025-12-31 announced 2025-12-23 |
Groq | playai-tts, playai-tts-arabic |
All self-serve tiers | canopylabs/orpheus-v1-english, orpheus-arabic-saudi | canopylabs/orpheus-3b-0.1-ft (Orpheus) |
| 2025-10-10 announced 2025-09-10 |
Groq | moonshotai/kimi-k2-instruct |
All self-serve tiers | moonshotai/kimi-k2-instruct-0905 (itself later retired) | — |
| 2025-10-08 announced 2025-08-08 |
Groq | gemma2-9b-it |
All self-serve tiers | llama-3.1-8b-instant (itself later retired) | — |
| 2025-10-02 announced 2025-09-02 |
Groq | deepseek-r1-distill-llama-70b |
All self-serve tiers | llama-3.3-70b-versatile (itself later retired) | — |
| 2025-08-30 announced 2025-05-31 |
Groq | llama3-70b-8192, llama3-8b-8192 |
All self-serve tiers | llama-3.3-70b-versatile, llama-3.1-8b-instant (both later retired) | Llama 3.3 70B |
How to use this page
- Your app started returning
404 model_not_found? Find the id above. If the shutdown date has passed, that is the cause — see the model_not_found entry for the fix ladder. - Planning ahead? Rows marked scheduled have not shut down yet. Move before the date, or add a fallback provider so the shutdown degrades gracefully.
- Want to keep the exact model? The last column links to it where it is still served. Inference APIs accepts the retired id as an alias in those cases, so only the base URL changes.
The pattern
Groq's list shows the cadence: roughly one retirement wave every one to two months, each announced about 30 days ahead by email and on the deprecations page, and several "replacements" that were themselves retired within the year (llama-3.1-8b-instant replaced gemma2-9b-it, then was retired; llama-3.3-70b-versatile replaced deepseek-r1-distill-llama-70b, then was retired). Pinning a model id without a fallback is a recurring outage, not a one-off.
Sources and corrections
Provider pages: Groq. Other providers will be added as their retirements are verified. Spotted a retirement we have not listed? Send the provider's announcement and we will add it.