fix: update Mistral model metadata for codestral-latest#703
Conversation
github-actions
Bot
requested review from
Alex Z (CLowbrow),
aswink,
Caitlin Pinn (cpinn),
Erin McNulty (erin2722) and
Ken Jiang (knjiang)
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0dceb0472c
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "Codex (@codex) review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".
| "displayName": "Codestral", | ||
| "max_input_tokens": 32000, | ||
| "max_output_tokens": 8191, | ||
| "max_input_tokens": 256000, |
There was a problem hiding this comment.
For codestral-latest, this raises the advertised input limit to 256k, but the current official Mistral docs for Codestral 25.08 list the context as 128k (see https://docs.mistral.ai/models/model-cards/codestral-25-08 and the model comparison page). Any UI or caller that relies on this catalog to size prompts can now send requests between 128k and 256k tokens that the Mistral API rejects, so the alias should be capped at 128000 unless there is a provider source for 256k.
Useful? React with 👍 / 👎.
2 participants
fix: update Mistral model metadata for codestral-latest
Closes #693
Source issue: #693
Summary
codestral-latestcodestral-latest2
Verified metadata
Verification notes
Verification
Official sources consulted
input_cost_per_mil_tokens,output_cost_per_mil_tokens.codestral-latestpoints tocodestral-2508. Verified fields: model alias relationship.codestral-2508entry inmodel_list.jsonalready hasinput_cost_per_mil_tokens: 0.3,output_cost_per_mil_tokens: 0.9,max_input_tokens: 256000, and listscodestral-latestas its parent. The-latestalias should match the snapshot it resolves to.sync_models (LiteLLM) cross-check
The relevant sync_models entry is
mistral/codestral-latest:input_cost_per_token: 1e-06 ($1.00/M) — stale, matches our current stale catalog valueoutput_cost_per_token: 3e-06 ($3.00/M) — stale, matches our current stale catalog valuemax_input_tokens: 32000 — stale, matches our current stale catalog valuemax_output_tokens: 8191 — stale, matches our current stale catalog valueAll four fields deviate from sync_models
mistral/codestral-latestbecause sync_models has also not updated this entry from the original May 2024 Codestral pricing. The official Mistral pricing page (https://mistral.ai/pricing) and the sync_modelsmistral/codestral-2508entry (input: 3e-07 = $0.30/M, output: 9e-07 = $0.90/M, max_input: 256000) both confirm the proposed values. The official source is preferred becausecodestral-latestresolves tocodestral-2508per Mistral's documentation.mistral/codestral-latest)input_cost_per_mil_tokensoutput_cost_per_mil_tokensmax_input_tokensmistral/codestral-2508agrees at 256000max_output_tokenscodestral-2508catalog entry omits max_output_tokens; no official source publishes a separate output limitFields not changed
format: remains"openai"(correct for Mistral provider)flavor: remains"chat"(correct for code generation chat model)displayName: remains"Codestral"(correct for a-latestalias)available_providers: remains["mistral"](correct)parent: not applicable —codestral-latestis the stable alias, not a snapshotdeprecated: not applicable — model is activesupported_regions: not applicable — not a Vertex modelsync_models vs proposed update
sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.