Mistral AI unveils Mistral 3 model family, including Mistral Large 3

Mistral AI has unveiled the Mistral 3 family, which includes three small dense models (14B, 8B, and 3B parameters) and the flagship Mistral Large 3, a sparse mixture-of-experts model trained with 41B active and 675B total parameters. All models are released under the Apache 2.0 license. Mistral Large 3 was trained from scratch on 3000 NVIDIA H200 GPUs and features image understanding and multilingual capabilities, achieving #2 among OSS non-reasoning models on the LMArena leaderboard. The Ministral 3 series offers 3B, 8B, and 14B parameter models with base, instruct, and reasoning variants, each supporting multimodal understanding. Mistral collaborated with NVIDIA, vLLM, and Red Hat to optimize inference and deployment. Mistral 3 is available on multiple platforms including Mistral AI Studio, Amazon Bedrock, Azure Foundry, Hugging Face, and others. Custom model training services are also offered for enterprise clients.