Mathstral
Mistral's 7B math specialist; 56.6% on MATH under Apache 2.0.
About
Released July 2024, Mathstral is Mistral AI's 7-billion-parameter model specialized for mathematical reasoning and scientific discovery, built in collaboration with Project Numina (the team behind the AIMO progress-prize-winning NuminaMath). At release it scored 56.6% on MATH and 77.1% on GSM8K, substantially ahead of same-size generalists like Llama 3 8B and competitive with much larger models of its era on competition math.
It is released under Apache 2.0 with open weights, making it one of the few permissively licensed math specialists — useful for embedding in tutoring products, STEM tooling, or as a verifier in math pipelines without license friction.
Limitations: it is a narrow specialist — general chat, coding, and instruction-following are weak compared to modern generalist models, and its 2024 training means it predates current reasoning-model techniques. AIME 2024 performance (2/30) shows it is not a competition-math solver by today's standards.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
56.6% MATH, 77.1% GSM8K, 37.2% Odyssey Math (vendor-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 64; Trending score 68/100 from researcher assessment.