Mistral Small 4 Review — 119B MoE, 256K Context, Reasoning + Vision + Coding in One

Mistral Small 4 (March 16, 2026) is a 119B Mixture-of-Experts model with 6.5B active parameters per token, 256K context, configurable reasoning, vision input, and Apache 2.0 license. GPQA Diamond 71.2%. MMLU-Pro 78.0%. HumanEval 92%. Outperforms GPT-OSS 120B on LiveCodeBench. $0.15/$0.60 per million tokens. Rating: 4/5.