Mistral releases Medium 3.5, a 128B open-weight multimodal model
Mistral Medium 3.5 is a dense text and vision model with a 256K-token context, available under a modified MIT license. It scored 77.6% on SWE-bench Verified and 91.4% on τ³-Telecom and consolidates Medium 3.1, Magistral, and Devstral 2. The API supports per-request reasoning effort and costs $1.50 per million input tokens and $7.50 per million output tokens. The model can be self-hosted on four GPUs.
Visit original source