llm-mistral 0.16 Brings Reasoning Model Support, Including Mistral Large 4

The llm-mistral plugin has released version 0.16, adding first-class support for reasoning models, with Mistral Large 4 as the headline addition. For anyone running Mistral models through Simon Willison's llm CLI, this is the direct path to the new reasoning variants without writing custom API glue.

Why it matters

Reasoning models are no longer a novelty confined to one or two providers. Mistral's entry into the space with Large 4 signals the pattern is now table stakes across the major API providers. The llm ecosystem is a de facto standard for prompt engineers and AI builders who want scriptable, composable model access. Bringing reasoning support into that ecosystem means the evaluation and prototyping loop for chain-of-thought workflows just got shorter.

The fastest way to evaluate a new reasoning model is through the toolchain you already trust.

What changes in practice

• Reasoning model access via CLI: Mistral Large 4 and any subsequent reasoning variants are now queryable through standard llm commands, no custom client code required.
• Unified comparison surface: Teams already benchmarking OpenAI o-series or other reasoning models through llm can now add Mistral to the same test harness.
• Plugin parity: The llm-mistral plugin catches up to the provider's current model lineup, closing the gap that often makes third-party plugins feel stale.

How to use it

1. Upgrade the plugin: llm install -U llm-mistral
2. Confirm the version: llm plugins should show llm-mistral 0.16.
3. Set your Mistral API key if you haven't: llm keys set mistral
4. Run a reasoning query: llm -m mistral-large-4 "Walk me through this step by step: [your prompt]"
5. Pipe it into your existing eval scripts or diff it against other reasoning models in the same session.

Check the release notes for the full list of supported model identifiers, as Mistral's naming conventions have shifted with this generation.

For teams already invested in the LLM Tooling ecosystem, 0.16 is a low-friction upgrade with immediate payoff for anyone evaluating reasoning models across providers.

One plugin update rarely changes your stack, but it does change which models you actually bother testing.