/
#56
Mistral AI
Paris-based lab founded by ex-DeepMind and Meta researchers. Releases open-weight LLMs under Apache 2.0, spanning dense transformers (Ministral 3B to 14B) and sparse MoE architectures (Mixtral, Mistral Large 3: 41B active / 675B total parameters, 256K context). Models target multilingual performance, edge deployment, agentic workflows, and coding. Inference-optimized for vLLM, TensorRT-LLM, and SGLang.
Links