Multiple issues with magistral medium

[LousyBook94]

App Version

3.26.0

API Provider

Mistral AI

Model Used

magistral-medium-latest

Roo Code Task Links (Optional)

https://app.roocode.com/share/7f0bc7d6-61a6-457b-8139-505374698043

🔁 Steps to Reproduce

1. Select a reasoning model from mistral which is magistral-medium
2. Send a message
3. The model's reasoning content is not formatted as reasoning content
4. The models full context (128k) is not available, a toggle to enable it would be great, as this slightly decreases perfomance
5. The model might trigger a context condense, which if there are a few messages, it will fail

💥 Outcome Summary

Expected the think content to be formatted, but instead got the full content in one big chunk, overloading the context a lot
Expected to have full context, but got the lesser context
Expected context condense to work, but didn't

📄 Relevant Logs or Errors (Optional)

[roomote]

I'm exploring a potential fix for the magistral-medium model issues (reasoning formatting, context limits, and condense failures) and will open a PR to share progress.

[roomote]

I've opened PR #7452 to address the magistral-medium model issues: #7452

The fix includes:

• ✅ Corrected context window from 41k to 128k tokens
• ✅ Added reasoning support flag for proper formatting
• ✅ All CI checks passing

These changes should resolve the reasoning content formatting, enable the full context window, and improve context condensing reliability.

[daniel-lxs]

Thank you @LousyBook94 for reporting this, I left my thoughts here: #7452 (comment)

I think it makes sense to have the two versions, I'll close this issue for now, feel free to open a new one to add the models or edit this one.