Mistral AI Open Models: Model List, Licenses and Articles
About Mistral AI
Mistral AI is a French AI company. Its mistralai organization on Hugging Face carries the tagline “Frontier AI. In Your Hands." and says new checkpoints will be released there. Mistral 7B, from 2023, became one of the standard 7B-class models for running locally.
What It Releases
- Mistral Small 4 (119B, 6.5B active) merges three families into one model. According to its model card, it unifies Instruct, Reasoning (previously Magistral) and Devstral in a single model. It is an MoE model with 4 of 128 experts active, a 256K-token context and image input.
- Mistral Medium 3.5 (128B dense) is “our first flagship merged model". It handles instruction following, reasoning and coding in one set of weights, and replaced Mistral Medium 3.1 and Magistral in Le Chat.
- The small Ministral 3 family (3B, 8B, 14B) is built for edge deployment. The 8B model card says it fits in 12GB of VRAM in FP8, and less if further quantized. It also reads images.
- Others include Voxtral for speech recognition, Leanstral, a code agent that writes Lean 4 proofs, and Shieldstral for safety classification.
Licenses
Apache-2.0 and a revenue-capped “Modified MIT License" are both in use. Mistral Small 4, Ministral 3 and Leanstral are Apache-2.0. The Modified MIT License used by Mistral Medium 3.5 and Devstral 2 (123B), however, says you may not exercise any rights under it if the global consolidated monthly revenue of your company (or your employer) exceeded US$20 million in the preceding month; in that case you request a commercial license from Mistral AI or use its hosted services. The Voxtral-4B-TTS speech synthesis model is non-commercial CC BY-NC 4.0. This summary is not legal advice; read each model’s license text before use.
Running Them Locally
- Official GGUF builds stop at Ministral 3. Mistral AI publishes GGUF files for the Ministral 3 3B, 8B and 14B models, but we find no official GGUF for Mistral Small 4 or Mistral Medium 3.5 (as of 2026-09-26).
- The large models have official NVFP4 builds and EAGLE heads for speculative decoding, meant to speed them up in vLLM and SGLang.
- For Mistral Medium 3.5, use GGUF files made after the configuration fix. The model card warns that the Transformers config had an error that degraded long-context performance, and that GGUFs generated from the config before the fix are affected too.
Sources: the mistralai organization card; model cards and license texts of mistralai/Mistral-Small-4-119B-2603, mistralai/Mistral-Medium-3.5-128B, mistralai/Ministral-3-8B-Instruct-2512 and mistralai/Devstral-2-123B-Instruct-2512 (all as of 2026-09-26).
Our Coverage and Data
Local Model Watch has published 0 article(s) on Mistral AI’s own models, 0 on third-party fine-tunes and quantized builds based on them, and 2 from its official blog. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.
At a Glance
| Item | Value |
|---|---|
| Hugging Face organization | mistralai |
| Models on Hugging Face | 75 |
| Articles on its own models | 0 |
| Articles on derived models | 0 |
| Articles from its official blog | 2 |
Recent Models on Hugging Face
The publisher lists 75 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.
→ Scroll horizontally to see all columns
| Released | Model | Task | Parameters | License | Official variants |
|---|---|---|---|---|---|
| 2026-07-16 | mistralai/Shieldstral-1.0-3B | — | 3.8B | apache-2.0 |
— |
| 2026-07-02 | mistralai/Leanstral-1.5-119B-A6B | — | — | apache-2.0 |
— |
| 2026-03-31 | mistralai/Mistral-Medium-3.5-128B | — | 127.7B | other |
EAGLE |
| 2026-03-11 | mistralai/Leanstral-2603 | — | — | apache-2.0 |
— |
| 2026-01-23 | mistralai/Mistral-Small-4-119B-2603 | — | 119.4B | apache-2.0 |
eagle, NVFP4 |
| 2026-01-22 | mistralai/Voxtral-Mini-4B-Realtime-2602 | audio | 4.4B | apache-2.0 |
— |
| 2025-11-30 | mistralai/Mistral-Large-3-675B-Base-2512 | — | — | apache-2.0 |
— |
| 2025-11-29 | mistralai/Mistral-Large-3-675B-Instruct-2512 | — | — | apache-2.0 |
NVFP4, Eagle, BF16 |
| 2025-11-29 | mistralai/Devstral-2-123B-Instruct-2512 | — | 125.0B | other |
— |
| 2025-11-28 | mistralai/Devstral-Small-2-24B-Instruct-2512 | — | 24.0B | apache-2.0 |
— |
| 2025-11-18 | mistralai/Voxtral-4B-TTS-2603 | audio | — | cc-by-nc-4.0 |
— |
| 2025-10-31 | mistralai/Ministral-3-3B-Instruct-2512 | — | 3.8B | apache-2.0 |
ONNX, GGUF, BF16 |
| 2025-10-31 | mistralai/Ministral-3-8B-Instruct-2512 | — | 8.9B | apache-2.0 |
GGUF, BF16 |
| 2025-10-31 | mistralai/Ministral-3-14B-Instruct-2512 | — | 13.9B | apache-2.0 |
GGUF, BF16 |
| 2025-10-31 | mistralai/Ministral-3-14B-Reasoning-2512 | — | 13.9B | apache-2.0 |
GGUF |
From the Hugging Face API. Last changed 2026-09-26 (JST). Collected by code.
From Its Official Blog
- 2026-09-10 Mistral AI and Cloudera Partner for On-Premise Sovereign AI
- 2026-09-10 Mistral AI Shares Fortran to C++ Legacy Migration via Agents
Last updated 2026-09-26 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.