Mistral Small 3.1 Review — 24B Vision Model, 128K Context, Runs on One GPU

Mistral Small 3.1 (March 17, 2025) added multimodal vision and a 4× context expansion (128K) to the 24B-parameter Mistral Small 3 base — while keeping the Apache 2.0 license and single-GPU deployment story. MMLU 80.62%. HumanEval 88.41%. DocVQA 94.08%. 150 tokens/second. $0.10/$0.30 per million tokens. Rating: 4/5.