“Magistral Medium scored 73.6% on AIME2024, and 90% with majority voting @64. Magistral Small scored 70.7% and 83.3% respectively.”
Mistral’s first reasoning model comes in two sizes, and only the smaller one has open weights. The headline 90% needs 64 tries and a vote. The pitch targets legal, finance, healthcare and government. Open source is the free sample.