Mistral Large 4 — Public Preview Open v26.10.
Architecture: granular Mixture-of-Experts; 52B active parameters, 1.05T total, 1.6B vision encoder. Natively multimodal.
API pricing (preview): 0.07/M cached input, $2.09/M output tokens. 1M context window.
Features: structured outputs, function calling, document QnA, agents & conversations, built-in tools, batching.
Model ID: mistral-large-4. Weights release end of October 2026.