# Mistral Small 4 — weights, license and what is actually open

A hybrid sparse model that combines instruction following, reasoning, visual understanding, and coding modes in one downloadable checkpoint.

Canonical: https://brightaifuture.com/open-models/mistral
Format: model-family
Source publication: Not established
Bright publication: 2026-09-19
Substantive update: None recorded
Evidence and review: AI-assisted primary-source check on 2026-09-19; no independent model reproduction or license certification. The registry records vendor evaluations but does not treat them as independent corroboration.

## Documented release

Mistral-Small-4-119B-2603

Organization: Mistral AI

Mistral Small 4 release: 2026-03-16 (day precision). Source: https://docs.mistral.ai/models/mistral-small-4-0-26-03

Parameters: 119B total / 6.5B active in Mistral’s model docs; its release article says 6B active per token (8B including embedding and output layers)

Architecture: Mixture-of-experts transformer with 128 experts and four active experts per token.

Modalities: text, image

Context: 256K tokens

## License and commercial use

Apache-2.0

Permitted by Apache-2.0.

## What is open

Weights: available. Official FP8 weights are downloadable. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Architecture: available. Architecture, expert counts, and configuration are published. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Inference code: available. The card documents vLLM and Transformers serving. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Training code: not-established. The full code used to train the model was not confirmed. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Training recipe: partial. The release describes capabilities and design without a full reproducible recipe. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Data information: not-established. A sufficiently detailed training-data account was not confirmed. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Training data: not-established. No sufficiently specific public artifact was confirmed in this review. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Evaluation: partial. Vendor results are published, including machine-readable evaluation artifacts. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

Commercial use: available. Apache-2.0 permits commercial use. Source: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603

## Uses and strengths described in sources

Reasoning mode

Visual understanding

Agentic coding

Efficient active parameter count

## Hardware and quantization

The FP8 repository is about 242 GB; the NVFP4 repository is about 71 GB before runtime overhead.

FP8

Official NVFP4 checkpoint

## Independent evidence

The registry records vendor evaluations but does not treat them as independent corroboration.

## Limitations

Training data is not documented in enough detail for reconstruction

Performance claims are primarily vendor-reported

## Provenance and history

{}

## Original sources

- [Introducing Mistral Small 4](https://mistral.ai/news/mistral-small-4/)
- [Mistral Small 4 model card](https://huggingface.co/mistralai/Mistral-Small-4-119B-2603)
- [Mistral Small 4 release](https://docs.mistral.ai/models/mistral-small-4-0-26-03)

## Continue exploring

- [Open Models](https://brightaifuture.com/open-models)
- [Open intelligence thread](https://brightaifuture.com/threads/open)