Mistral dropped two reasoning models: 24B param, open-source Magistral Small and enterprise-specific Magistral Medium

Both models' chain-of-thought works across global languages and alphabets

However, STEM and coding benchmarks lag behind top rivalspic.twitter.com/nfqZQpBlWw

— Rowan Cheung (@rowancheung) 11 juin 2025

Mistral dropped two reasoning models: 24B param, open-source Magistral Small and enterprise-specific Magistral Medium Both models' chain-of-thought works across global languages and alphabets However, STEM and coding benchmarks lag behind top rivals

→ View original post on X — @rowancheung