Atlas

Models

← All models

Magistral Medium 1.0

Mistral AI · 2025-06-10

Magistral Medium 1.0 is Mistral AI's first frontier-class reasoning model, an enterprise model released alongside open-weight Magistral Small 1.0 in June 2025. It was trained on top of Mistral Medium 3 using reinforcement learning alone, always generates a thinking trace, and targets multilingual, multi-step reasoning across mathematics, coding, STEM, and professional use cases. Mistral retired it on November 30, 2025 and lists Mistral Medium 3.5 as its replacement.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AA-LCR v1.10.0
AA-Omniscience Index-26.9
AIME 2025 (Artificial Analysis avg@10)40.3
ARC-AGI-1 Semi-Private6.1
ARC-AGI-20.0
Arena Text — Business, Management & Financial Ops (No Style Control)1223
Arena Text — Chinese (No Style Control)1228
Arena Text — Creative Writing (No Style Control)1245
Arena Text — English (No Style Control)1274
Arena Text — Entertainment, Sports & Media (No Style Control)1241
Arena Text — Exclude Ties (No Style Control)1177
Arena Text — Exclude Ties (Style Controlled)1249
Arena Text — Expert (No Style Control)1223
Arena Text — French (No Style Control)1267
Arena Text — French (Style Controlled)1325
Arena Text — German (No Style Control)1248
Arena Text — Hard Prompts (No Style Control)1266
Arena Text — Hard Prompts English (No Style Control)1284
Arena Text — Instruction Following (No Style Control)1254
Arena Text — Japanese (No Style Control)1177
Arena Text — Korean (No Style Control)1125
Arena Text — Legal & Government (No Style Control)1232
Arena Text — Life, Physical & Social Science (No Style Control)1233
Arena Text — Longer Query (No Style Control)1294
Arena Text — Medicine & Healthcare (No Style Control)1245
Arena Text — Multi-turn (No Style Control)1275
Arena Text — Non-English (No Style Control)1232
Arena Text — Non-English (Style Controlled)1281
Arena Text — Occupational Mathematical (No Style Control)1256
Loading Atlas data…