Mistral AI unveils Magistral 1.2 models with enhanced multimodal and benchmark performance

Mistral AI has released Magistral Small 1.2 and Medium 1.2 models, now featuring multimodal capabilities that support both text and image inputs through a vision encoder. These updates bring a 15% improvement on key math and coding benchmarks, including AIME 24/25 and LiveCodeBench.

Tool use for web search, code interpretation, and image generation is also enhanced.