Partially Verified75% confidenceFactExact time
Mistral Large 4 adopts a sparse Mixture of Experts (MoE) architecture, with 49 billion (49B) parameters actually activated during each inference.
3
Sources
75%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedMistral released Mistral Large 4, codenamed LeChunk, a trillion-parameter, natively multimodal MoE model with approximately 49 billion active parameters.80% similarUnverifiedBased on the repository naming convention A52B, it is speculated that Mistral Large 4 activates 52 billion parameters per forward pass, though an official full specification has not yet been released.76% similarUnverifiedMistral Large 4's vision encoder has been reduced from 2.5 billion parameters in the previous generation to 1.6 billion parameters.76% similarUnverifiedMistral Large 4 scored approximately 62% on the real-world software engineering benchmark DeepSwee.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/995373API
curl https://kongchang.com/api/v1/knowledge/claims/995373MCP
get_claim(id=995373)