mistral 3 ships apache 2.0 open weights from 3b to 675b parameters

Mistral released Mistral Large 3 and three Ministral 3 sizes under Apache 2.0, on its own platform, Hugging Face and the major clouds.

2025-12-02 ai 2 sources · 2 primary 1 min read

// at a glance

version: Mistral Large 3; Ministral 3 (14B, 8B, 3B)
platforms: Mistral AI Studio, Hugging Face, Amazon Bedrock, Azure Foundry, IBM watsonx, OpenRouter, Fireworks, Together AI

what happened

Mistral announced Mistral 3 on December 2, 2025, stating: 'All models are released under the Apache 2.0 license.'[1]
Mistral Large 3 is a sparse mixture-of-experts model with 41B active and 675B total parameters, released in base and instruct versions.[1]
Ministral 3 comes in dense 14B, 8B and 3B sizes, each with base, instruct and reasoning variants and image understanding.[1]
The models are available in Mistral AI Studio and on Hugging Face, Amazon Bedrock, Azure Foundry, Modal, IBM watsonx, OpenRouter, Fireworks, Unsloth AI and Together AI.[1]
Mistral lists API pricing of $0.5 input and $1.5 output per million tokens for Large 3, and $0.15 / $0.15 for Ministral 3 8B.[1]
Mistral followed on October 6, 2026 with Mistral Large 4, a 1-trillion-parameter model with 52B active parameters, in public preview on Mistral Studio, saying it would release the weights by the end of the month.[2]