Mistral Becomes Inference Provider for Chinese Open Models, Flagship Architecture Mirrors DeepSeek

SumitGup · x · 2026-08-14

According to a repost, Mistral is becoming an inference provider for open-source Chinese models. The post notes that Mistral Large 3 (December 2025 flagship) is built on DeepSeek V3's architecture with nearly identical config (hidden dim 7168, 61 layers, 3 dense layers, etc.), differing only in MoE granularity. Mistral raised over $3 billion but has shifted strategy.

Original post →

More from Companies & People

Companies & People channel →