[ENHANCEMENT] Mistral AI Devstral 2 Support

[machinekoder]

Problem (one or two sentences)

Devstral 2 was just released: https://mistral.ai/news/devstral-2-vibe-cli

The Mistral AI Provider should be updated to support the new Devstral 2 models

Context (who is affected and when)

Mistral AI Provider

Desired behavior (conceptual, not technical)

Devstral 2 Models selectable and preconfigured:

Devstral 2: https://docs.mistral.ai/models/devstral-2-25-12

names: devstral-latest, devstral-medium-latest, devstral-2512
price: input 0.4/M tokens, output 2.0/M tokens
context window: 256k

Devstral Small 2: https://docs.mistral.ai/models/devstral-small-2-25-12

names: labs-devstral-small-2512, devstral-small-latest
price: input 0.1/M tokens, output 0.3/M tokens
context window: 256k

Constraints / preferences (optional)

No response

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Roo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

No response

[roomote]

I've analyzed this issue and here's my proposed implementation plan:

Implementation Plan for Devstral 2 Support

I'll add support for the new Devstral 2 models by updating the Mistral provider configuration:

1. Update Model Definitions

In packages/types/src/providers/mistral.ts, I'll add the following new models to the mistralModels object:

Devstral 2 (devstral-2512):

• Model IDs: devstral-latest, devstral-medium-latest, devstral-2512
• Context window: 256k tokens
• Pricing: /bin/sh.4/M input, .0/M output
• Features: Native tools support, images support

Devstral Small 2 (labs-devstral-small-2512):

• Model IDs: labs-devstral-small-2512, devstral-small-latest
• Context window: 256k tokens
• Pricing: /bin/sh.1/M input, /bin/sh.3/M output
• Features: Native tools support

2. Model Configuration

I'll configure each model with:

• maxTokens: 8192 (standard for Mistral models)
• contextWindow: 256_000 (256k as specified)
• supportsImages: true for Devstral 2, false for Devstral Small 2
• supportsPromptCache: false (standard for Mistral)
• supportsNativeTools: true (both models support tools)
• Input/output pricing as specified in the issue

3. Update Existing Model References

• The existing devstral-medium-latest entry will be updated to match the new Devstral 2 specifications
• I'll ensure backward compatibility by keeping the existing model ID

4. Testing

• Update existing tests to verify the new models are properly configured
• Ensure the provider correctly routes requests to the appropriate endpoint

This approach will:

• Add full support for both new Devstral 2 models
• Maintain backward compatibility with existing configurations
• Follow the established pattern for Mistral model definitions
• Properly configure pricing and capabilities as specified

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[machinekoder]

@roomote I updated the cost from the original issue text

both models support image input

It might be worth exploring the Mistral Vibe CLI code such as the system prompt for optimal utilization: https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/system_prompt.py

Note also how the models are configured in Vibe CLI https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/config.py#L272-L294

also note "For optimal performance, we recommend a temperature of 0.2 and following the best practices defined for Mistral Vibe CLI."

[roomote]

Thanks for the updated information @machinekoder! I see the corrections about both models supporting image input and the recommendations about exploring the Mistral Vibe CLI configuration. I'll incorporate these insights into the implementation, including:

• Configuring both models with image support
• Reviewing the Vibe CLI system prompt and configuration patterns
• Considering the recommended temperature of 0.2 for optimal performance

I'll explore the codebase and create a PR to address this enhancement.

[machinekoder]

@roomote sounds good, go ahead

[roomote]

I've seen your comment and will explore a potential fix. I'll open a PR to share progress.

[roomote]

I've opened a draft PR with my attempt at addressing this issue: #9957

The implementation includes:

• Added all four new Devstral 2 model variants as specified
• Updated both models to support image input as you noted
• Changed the default temperature to 0.2 following the Vibe CLI recommendations
• Updated the default model to use the new Devstral 2

All tests pass and the changes maintain backward compatibility. Feedback welcome!