Update mistral-medium prices and context sizes

[regismesquita]

trafficstars

While testing the Mistral model, I noticed a discrepancy in the pricing shown on the logs screen. After reviewing the code, I confirmed that the pricing values were incorrect.

This PR corrects the input and output token pricing for the latest Mistral model and adds the newly released mistral-medium-2505 version.

Title

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• [ ] I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• [ ] I have added a screenshot of my new test passing locally
• [ ] My PR passes all unit tests on make test-unit
• [ ] My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🆕 New Feature 🐛 Bug Fix 🧹 Refactoring 📖 Documentation 🚄 Infrastructure ✅ Test

Changes

[vercel[bot]]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 27, 2025 9:56pm

[CLAassistant]

All committers have signed the CLA.

[regismesquita]

@emerzon Mistral medium latest is only 32k?! Every place that I checked shows 128k.

[regismesquita]

Does mistral/mistral-medium points to the old one?

[emerzon]

@regismesquita It's probably better to add a new entry for the new model matching the API naming from mistral: mistral-medium-2505 and mistral-medium-latest to avoid removing information from the past model. https://docs.mistral.ai/getting-started/models/models_overview/

Also, mind that the the output context size (max_tokens and max_output_tokens) is not 128k - No clear source but probably still 8k.

[regismesquita]

This very link that you shared shows max tokens at 128k. I know they might be actually talking about the input context there... but going by the word... I have also found sources for output at 128k but none reliable , only stuff like openrouter listing of the new model.

I will remove mistral-medium name from the scope of this PR and retain only the 2503 and the latest identifiers.

I will try to experiment with this model today and also try to contact mistral.

[regismesquita]

@emerzon I have reverted mistral/mistral-medium to the old values, and now only the -latest alias and the -2503 aliases are updated/created, I am keeping the 128k value for now while I figure this out, I have asked around and I am trying to confirm the actual values, if I can't find any I will do exploratory requests later today.

[emerzon]

128k is for input tokens - not output tokens.

[regismesquita]

@emerzon can you share how are you sure of that? I couldn't confirm that anywhere in their docs.

[regismesquita]

@emerzon haven't got any response yet, and I do think that it is less than 128k but since I haven't got any response from them I prefer to err on the safe side. so I changed it down to 8191 , and I can follow up later if I found any evidence against it.

[regismesquita]

Ignore this comment.

[regismesquita]

synced with the upstream.