Mistral Small models

Strongest: Mistral Small 4

Mistral Small 4 (release 26.03, shipped 16 March 2026) is the cost-disruption release of the 2026 lineup: a 119B-parameter Mixture-of-Experts with only 6.5B active per token (128 experts, 4 active), priced at $0.15/$0.60 per 1M tokens, under genuine Apache 2.0. Despite the "Small" label it scores 78.0% on MMLU-Pro and 71.2% on GPQA Diamond — within reach of Medium 3.1 and Large 3 — while running on a single GPU. It unifies chat, reasoning, and coding in one efficient model. The buyer's sentence: the best price-to-capability open multimodal model in 2026, with Apache 2.0 as a structural moat. - Provider: Mistral AI (Paris, France) - Release: 2026-03-16, status GA - Context: 256,000 tokens; max output 16,384 - Modalities: text + image in, text out (native multimodal) - Knowledge cutoff: ~January 2026 - Headline price: $0.15 input / $0.60 output per 1M tokens - Architecture: MoE, 119B total / 6.5B active, 128 experts (4 active), MLA attention