backend deprecated models check - 2026-09-11

[reynaldichernando]

Automated check of the backend's hardcoded model lists against the live model API (https://ai-models.puter.work/<provider>).

Reports only deprecated models — backend entries whose id and none of whose aliases appear anywhere in the upstream name set (upstream id + name + every aliases entry, lowercased, models/ prefix stripped). Read-only; no files were modified.

Repo commit checked: 9940beb

Deprecations

chat/openai (1)

• o3-pro

  Notes: upstream still serves o3, o3-mini, o3-2025-04-16, o4-mini — only the -pro variant is gone. Backend aliases o3-pro-2025-06-10 and openai/o3-pro are also absent upstream.

chat/mistral (2)

• mistral-large-2512

• zai-glm-5-2

  Notes: the Mistral endpoint's full model list has no mistral-large* entry at all (mistral-large-latest doesn't resolve either) — only small/medium/ministral/codestral/magistral/voxtral remain. zai-glm-5-2 (glm-5-2), a third-party GLM model hosted on Mistral's infra, also has no upstream entry.

chat/deepseek (2)

• deepseek-v4-flash

• deepseek-v4-flash-vision-exp

  Notes: the DeepSeek endpoint now lists only deepseek-flash and deepseek-v4-pro. Neither backend model's id nor any of its aliases (deepseek-chat, deepseek-reasoner, etc.) matches upstream — deepseek-v4-pro itself is unaffected and still resolves.

tts/xai (1)

• xai-tts

  Notes: this is the only entry in ai-tts/providers/xai/costs.ts, so the whole xAI TTS provider is affected. The xAI endpoint does cover non-chat modalities (it lists grok-imagine-image*, matching the backend's image models), but contains no tts/speech/audio/voice model under any id or alias. Same as prior weeks' finding — worth a manual confirmation before removal, since xAI may simply not expose TTS on this listing.

Clean (no deprecations)

chat: claude, gemini, moonshot, minimax, xai, zai, alibaba  ·  image: gemini, openai, xai  ·  tts: gemini, openai  ·  video: gemini

(No ai-video provider exists for openai in this repo, so that pair was not applicable to check.)

Verify manually

zai: none — the endpoint returned its full 10-model catalog (glm-5.3, glm-5.3-flash, glm-5.2, glm-5.1, glm-5, glm-5-turbo, glm-4.7, glm-4.6, glm-4.5, glm-4.5-air) and every backend zai model matched directly this run.

[reynaldichernando]

sharing context what this automated check is about @404oops
every week we pull the official ai model list data from all our providers, then compare the data to what we have in our backend

if any model is missing from the official provider list, but we still have it in our backend, it means that the model is removed
and we'll have to remove it from our model.ts

but providers usually don't show all their available models, that's why we still have to verify it by calling the AI manually

for example with o3-pro

[linear]

from filip.kujundzic:

Yeah, I understand, though why not check for added models as well? seems like only checking for deprecations is counterintuitive given the velocity at which models keep arriving

[linear]

from reynaldi.chernando:

actually this is a different problem
this deprecated check was set up because its hard to know when models disappear
they get discontinued but we still show them in our model list (https://api.puter.com/puterai/chat/models/), misleading developers thinking they're still available
that's why we have this automation

as for new models, yeah we can add it, though this is less of an issue since we will at least always have openrouter or infron covering us