Mixtral 8x7B

Mixtral 8x7B — 46.7B total / 12.9B per token, Apache 2.0

Why it matters

Made Sparse MoE a standard architecture of the open ecosystem and imprinted the concept of total versus active parameters

Source — Primary sources first; secondary reports are labelled “Press”

Officialmistral.aihttps://mistral.ai/news/mixtral-of-experts

Added in Week 35, 2026 · See that week

Same week

Weak-to-strong generalizationOfficialUpstage releases SOLAR 10.7BOfficialSGLang paperPaper

Nearby on the same axis

Upstage releases SOLAR 10.7BOfficialDeepSeek releases DeepSeek LLM 7B and 67BOfficialGemma 1 (2B/7B), under Gemma's own terms of usePressMistral 7BOfficial