Mistral Large 3 launches under Apache 2.0 — 200B mixture-of-experts model ships with reasoning-mode variant, commercial-deployment terms unchanged from prior open releases
Mistral AI launched Mistral Large 3 on May 28 — a 200B mixture-of-experts model under Apache 2.0 terms, with a reasoning-mode variant trained on the company's preference-optimization stack. The Apache 2.0 licensing keeps commercial-deployment terms unchanged from prior Mistral open releases, and the model lands as the European open-source counterpart to Meta's Llama 4.1 70B Thinking release the same day.
The model architecture is the substantive piece. Mistral Large 3 uses a 200B-parameter mixture-of-experts design with ~36B active parameters per forward pass, sitting between the Llama 4.1 70B dense baseline and the larger Llama 4.1 405B dense option on the open-weight frontier. The reasoning-mode variant is trained on a combination of synthetic reasoning traces and the company's preference-optimization stack — Mistral has not disclosed the specific RLHF or DPO methodology, but the model's evaluation results on the reasoning benchmark suite are competitive with the closed-frontier-lab tier. The Apache 2.0 license terms are unchanged from prior Mistral releases — full commercial deployment, no acceptable-use restrictions, no derivative-work limitations.
The competitive context is the European-open-source positioning. Meta's Llama 4.1 70B Thinking is the US open-weight reasoning-mode release; Mistral Large 3 is the European counterpart. The two releases collectively establish that the open-weight ecosystem has reasoning-mode-trained models competitive with the closed-frontier-lab tier — a structural shift that was not credible a year ago. For European enterprises and government deployments where data-residency or sovereignty requirements drive the open-source-over-closed-API decision, Mistral Large 3's Apache 2.0 terms and EU-based provenance are the procurement-relevant differentiators. Cohere's earlier Aleph Alpha acquisition consolidated one part of the European open-source landscape; Mistral's continued cadence is the second pillar.
Mistral AI — Mistral Large 3 Apache 2.0 launch May 28 2026 → · Hugging Face — Mistral Large 3 200B MoE model weights → · Le Monde — Mistral Large 3 European open-source AI strategy →