Mistral Inference v1.6.0 Adds Small 3.1 Model Support

Mistral AI has released version 1.6.0 of its inference library, adding support for the Mistral Small 3.1 model with vision capabilities. The GitHub release shipped March 20th with model updates and code maintenance fixes.

The mistral-inference library handles local deployment and inference for Mistral’s language models. This update brings the Small 3.1 model into the supported lineup, expanding options for developers running Mistral models in their own infrastructure.

Mistral Small 3.1 Support Added

The headline feature is native support for Mistral Small 3.1, which includes vision capabilities alongside text processing. This means developers can now run the multimodal Small 3.1 model locally through the inference library, rather than relying solely on Mistral’s hosted API.

Small 3.1 sits in Mistral’s mid-tier model range — more capable than the base models but lighter than the large variants. The vision support lets it process images alongside text prompts, useful for applications that need to analyze visual content or combine text and image understanding.

Code Cleanup and Maintenance

The release includes several maintenance improvements. A missing newline bug was fixed, and the team removed outdated file references that were cluttering the codebase. These changes don’t affect functionality but clean up the developer experience.

The library now has a more streamlined structure after removing deprecated references, making it easier for contributors to navigate and maintain going forward.

New Contributors Join

Two new contributors made their first commits to the project. Theophile Gervet fixed the newline formatting issue, while Julien Denize handled the Small 3.1 integration and file cleanup work.

Open source contributions like these help expand Mistral’s inference capabilities while distributing maintenance work across a broader developer community.

Bottom Line

This is a solid maintenance release that adds a useful model to the local inference options. The Small 3.1 support matters most for developers who want vision capabilities without running the larger, more resource-intensive models. The code cleanup suggests active maintenance, which is always good to see in inference libraries that need to stay current with model releases.