Tier 2 Restricted open-weight
What the publisher claims: open-weight, open models
The one-liner: European frontier-scale open weights with a commercial carve-out.
The six questions
Weights were published on Hugging Face as Mistral-Large-4-1T-A52B on 29 October 2026, in FP8 and NVFP4.
The Ecosystem Open-Weight license is a modified MIT whose single condition bites a whole business model: any Model-as-a-Service operation must sign a separate commercial license with Mistral before commercial use, which is a meaningful restriction on redistribution as a service.
Self-hostable on large clusters with vLLM and SGLang support and third-party hosting from launch, with 1M-token context.
A model card, launch post, and benchmark results are published; no training data or reproducible training recipe accompanies the weights.
Limited: the release documents the model, not the process that produced it.
Large 4 ships weights, a model card, and serving guidance with no accompanying code repository or contribution channel for this checkpoint, so the only participation path is filing issues on Mistral's own pages.
License
Mistral Ecosystem Open-Weight License, a modified MIT — license source ↗
- Businesses that operate a Model-as-a-Service offering must obtain a separate commercial license from Mistral before using the model commercially
- Attribution notice and license text must be retained in copies and substantial portions
What is actually released
- weights (1.05T total, 52B active, 1M context; FP8 and NVFP4)
- model card
- deployment guidance for vLLM and SGLang
- evaluation results
Why it matters
Large 4 marks Mistral's flagship moving off Apache 2.0, and its MaaS carve-out joins the 2026 pattern of vendor-set revenue and business-model conditions attached to nominally open weights.
Sources
- https://mistral.ai/news/mistral-large-4-oss
- https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B
- https://www.datastudios.org/post/mistral-large-4-open-weights-1m-context-modified-mit-license
- https://mlq.ai/news/mistral-releases-1-trillion-parameter-large-4-model-under-a-restricted-open-weight-license
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.