⌘K
Change language Switch ThemeSign In
Narrow Mode
Introducing Mistral Small 4 ===========================
S Simon Willison's Weblog @Simon Willison
One Sentence Summary
Simon Willison introduces Mistral Small 4, a new 119B parameter MoE model that unifies reasoning, multimodal, and coding capabilities under an Apache 2 license.
Summary
The article reports on the release of Mistral Small 4, a significant 119B parameter Mixture-of-Experts (MoE) model from Mistral AI. This model is notable for unifying the specialized capabilities of Mistral's previous flagship models—Magistral (reasoning), Pixtral (multimodal), and Devstral (coding)—into a single versatile architecture. It supports adjustable reasoning effort levels and is released under the Apache 2 license. The author provides a brief practical demonstration using the 'llm' CLI tool and mentions the simultaneous release of Leanstral, a model specifically optimized for the Lean 4 formal verification language.
Main Points
* 1. Mistral Small 4 unifies specialized model capabilities into one versatile 119B MoE model.The model integrates reasoning, multimodal, and agentic coding features previously found in separate flagship models like Magistral and Pixtral, creating a more general-purpose tool. * 2. The model is released under the Apache 2 license, promoting open accessibility.Despite its large size (242GB on Hugging Face), the permissive licensing allows for broader community use and integration into various developer workflows. * 3. Mistral is targeting niche technical domains with specialized releases like Leanstral.Alongside the general model, Mistral released Leanstral, specifically tuned for the Lean 4 formally verifiable coding language, showing a trend toward domain-specific optimization.
Metadata
AI Score
85
Website simonwillison.net
Published At Yesterday
Length 200 words (about 1 min)
Sign in to use highlight and note-taking features for a better reading experience. Sign in now
16th March 2026 - Link Blog Introducing Mistral Small 4. Big new release from Mistral today (despite the name) - a new Apache 2 licensed 119B parameter (Mixture-of-Experts, 6B active) model which they describe like this:
> Mistral Small 4 is the first Mistral model to unify the capabilities of our flagship models, Magistral for reasoning, Pixtral for multimodal, and Devstral for agentic coding, into a single, versatile model.
It supports reasoning_effort="none" or reasoning_effort="high", with the latter providing "equivalent verbosity to previous Magistral models".
The new model is 242GB on Hugging Face.
I tried it out via the Mistral API using llm-mistral:
llm install llm-mistral
llm mistral refresh
llm -m mistral/mistral-small-2603 "Generate an SVG of a pelican riding a bicycle"
I couldn't find a way to set the reasoning effort in their API documentation, so hopefully that's a feature which will land soon.
Also from Mistral today and fitting their -stral naming convention is Leanstral, an open weight model that is specifically tuned to help output the Lean 4 formally verifiable coding language. I haven't explored Lean at all so I have no way to credibly evaluate this, but it's interesting to see them target one specific language in this way.
S Simon Willison's Weblog @Simon Willison
One Sentence Summary
Simon Willison introduces Mistral Small 4, a new 119B parameter MoE model that unifies reasoning, multimodal, and coding capabilities under an Apache 2 license.
Summary
The article reports on the release of Mistral Small 4, a significant 119B parameter Mixture-of-Experts (MoE) model from Mistral AI. This model is notable for unifying the specialized capabilities of Mistral's previous flagship models—Magistral (reasoning), Pixtral (multimodal), and Devstral (coding)—into a single versatile architecture. It supports adjustable reasoning effort levels and is released under the Apache 2 license. The author provides a brief practical demonstration using the 'llm' CLI tool and mentions the simultaneous release of Leanstral, a model specifically optimized for the Lean 4 formal verification language.
Main Points
* 1. Mistral Small 4 unifies specialized model capabilities into one versatile 119B MoE model.
The model integrates reasoning, multimodal, and agentic coding features previously found in separate flagship models like Magistral and Pixtral, creating a more general-purpose tool.
* 2. The model is released under the Apache 2 license, promoting open accessibility.
Despite its large size (242GB on Hugging Face), the permissive licensing allows for broader community use and integration into various developer workflows.
* 3. Mistral is targeting niche technical domains with specialized releases like Leanstral.
Alongside the general model, Mistral released Leanstral, specifically tuned for the Lean 4 formally verifiable coding language, showing a trend toward domain-specific optimization.
Key Quotes
* Mistral Small 4 is the first Mistral model to unify the capabilities of our flagship models, Magistral for reasoning, Pixtral for multimodal, and Devstral for agentic coding, into a single, versatile model. * It supports reasoning_effort='none' or reasoning_effort='high', with the latter providing 'equivalent verbosity to previous Magistral models'. * Leanstral, an open weight model that is specifically tuned to help output the Lean 4 formally verifiable coding language.
AI Score
85
Website simonwillison.net
Published At Yesterday
Length 200 words (about 1 min)
Tags
Mistral AI
Large Language Models
Open Source
Mixture-of-Experts
Lean 4
Related Articles
* Anthropic Introduces Claude Opus 4.6 with 1M Token Context * My fireside chat about agentic engineering at the Pragmatic Summit * OpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491 * Introducing Showboat and Rodney， so agents can demo what they’ve built * ClawdBot is out of control * 2025: The year in LLMs * Anthropic Releases Claude Sonnet 4.6 with 1M Context Window * Wilson Lin on FastRender: a browser built by thousands of parallel agents * First impressions of Claude Cowork， Anthropic’s general agent * NVIDIA’s Jensen Huang on Reasoning Models， Robotics， and Refuting the “AI Bubble” Narrative HomeArticlesPodcastsVideosTweets
Introducing Mistral Small 4 | BestBlogs.dev ===============