Mistral Small 3.2 24B: Open-Weight Model Update Ships Improved Function Calling, Repetition Reduction, Arena Hard Gains

VentureBeat·medium signal

Mistral updated Small 24B from 3.1 to 3.2 in March 2026, targeting instruction-following accuracy, reduced infinite generations, and measurable gains on WildBench and Arena Hard benchmarks. Function calling and structured output accuracy improved significantly over 3.1. The model remains fully open-weight with 128K context and multimodal (image + text) input support.