Mistral Medium 3.5 Review: 128B Dense, 77.6% SWE-Bench, and a Three-Model Consolidation

Mistral Medium 3.5 (April 29, 2026) is Mistral AI's most capable open-weights model to date: 128B dense parameters, 256K context, multimodal, and adjustable reasoning built in. It replaces three prior Mistral models — Medium 3.1, Magistral, and Devstral 2 — with a single API endpoint. SWE-Bench Verified 77.6%. $1.50/$7.50 per million tokens. Modified MIT license with a $20M/month revenue cap. The consolidation is real. So is the price increase. Rating: 3.5/5.