Mistral Large 3

weights

open(Apache-2.0)

data

closed

code

closed

license

Apache-2.0(OSI)

Mistral Large 3 ships Apache-2.0 weights, but the training data is closed and no training code is published. Mistral relicensed between releases: Large 2 shipped under the Mistral Research License, permitting research and non-commercial use only, while Large 3, the current release, is Apache-2.0.

• https://azure.microsoft.com/en-us/blog/introducing-mistral-large-3-in-microsoft-foundry-open-capable-and-ready-for-production-workloads/ recorded 2026-09-29

  Microsoft's Foundry launch post calls it "one of the strongest open-weight, Apache-licensed frontier models" under the heading "Fully Open and Apache 2.0 licensed". A distribution partner, corroborating only.

• https://docs.mistral.ai/models/mistral-large-3-25-12 recorded 2026-09-29

  Mistral's model reference describes "Mistral Large 3, is a state-of-the-art, open-weight, general-purpose multimodal model" and lists the license as Apache 2.0.

• https://huggingface.co/api/models/mistralai/Mistral-Large-3-675B-Base-2512 recorded 2026-09-29

  Hugging Face API record for "mistralai/Mistral-Large-3-675B-Base-2512" carries the tag "license:apache-2.0" and reports gated as false and private as false. Searched the body for agree to share and access request: none.

• https://huggingface.co/mistralai/Mistral-Large-3-675B-Base-2512/raw/main/README.md recorded 2026-09-29

  The base model card front matter declares "license: apache-2.0" and the card states "This model is the base pre-trained version, not fine-tuned for instruction or reasoning tasks" with an Apache 2.0 License entry reading "Open-source license allowing usage and modification for both commercial and non-commercial purposes". Searched the card for dataset or training-code links: none.

• https://mistral.ai/news/mistral-3/ recorded 2026-09-29

  Mistral's launch post: "We release both the base and instruction fine-tuned versions of Mistral Large 3 under the Apache 2.0 license" and "trained from scratch on 3000 of NVIDIA". Searched the body for training data and a released dataset or training-code repository: none, so the corpus and pipeline stay closed.

Adoption is measured as Hugging Face downloads summed across the three declared artifacts, spanning the retired Mistral Large 2 and the current Large 3 base and instruct checkpoints. The flagship is mostly consumed through Mistral's API rather than downloaded, so this understates real use, but only the artifacts a product declares count toward this reading.

• https://huggingface.co/api/models/mistralai/Mistral-Large-3-675B-Base-2512 recorded 2026-08-12

  80 downloads in the trailing 30 days for mistralai/Mistral-Large-3-675B-Base-2512

• https://huggingface.co/api/models/mistralai/Mistral-Large-3-675B-Instruct-2512 recorded 2026-08-12

  1,177 downloads in the trailing 30 days for mistralai/Mistral-Large-3-675B-Instruct-2512

• https://huggingface.co/api/models/mistralai/Mistral-Large-Instruct-2407 recorded 2026-08-12

  4,714 downloads in the trailing 30 days for mistralai/Mistral-Large-Instruct-2407

Mistral Large 3 scores 9 on Artificial Analysis's Intelligence Index, below the median of 12 for open-weight non-reasoning models of similar size. Mistral's launch post placed it second among open non-reasoning models on LMArena, but that was a launch-day placement rather than a current standing.

• https://artificialanalysis.ai/models/mistral-large-3 recorded 2026-09-29

  Artificial Analysis model page on Intelligence Index v4.3.2: "Mistral Large 3 scores 9 on the Artificial Analysis Intelligence Index, placing it below average among other open weight non-reasoning models of similar size (median: 12)".

• https://mistral.ai/news/mistral-3/ recorded 2026-09-29

  Mistral's launch post: "Mistral Large 3 debuts at #2 in the OSS non-reasoning models category (#6 amongst OSS models overall)" on LMArena. A launch-day placement.