Mistral Medium 3

Mistral Medium 3 is Mistral AI's May 2025 frontier-mid model: a multimodal (text + image) model with tool use, structured outputs and a 131K context window, aimed at enterprise coding and STEM workloads. Mistral positions it as delivering ~90% of Claude Sonnet 3.7's performance at roughly 8x lower cost ($0.40/$2.00 per 1M input/output tokens), citing its own academic benchmarks plus third-party human evaluations of real-world use cases. It is available on Mistral's La Plateforme API, OpenRouter and cloud marketplaces, and was followed by Medium 3.1 (multimodal improvements) and Medium 3.5. Closed weights; Mistral only open-sources its Small/Mistral-series lines.

Provider-reported (mistral.ai/news/mistral-medium-3): ~90% of Claude Sonnet 3.7 performance at ~8x lower cost; Mistral highlights coding and STEM strengths; page cites third-party human evaluations of real-world use — all numbers from Mistral's internal evaluation pipeline.

Independent: Artificial Analysis lists the model but flags its Intelligence Index (~9.0) as estimated with no live evaluations run — i.e. no strong independent eval anchor. Aggregator coverage (galaxy.ai, magica.com comparisons) repeats the pricing/context specs but adds no independent benchmark runs.