Mistral has launched Medium 3.5, a 128-billion parameter model with a 256K token context window. The model weights are available on Hugging Face under a modified MIT license, and the API pricing is $1.50/$7.50 per million input/generated tokens. Medium 3.5 achieves a reasoning score of 77.6% on SWE-Bench Verified, outperforming Claude Sonnet 4.5.

Alongside the model, Mistral introduced Remote Agents in the Vibe environment, allowing code refactoring or generation sessions started locally to continue asynchronously in the cloud. These agents operate in isolated sandboxes, manage dependencies, make edits, and create pull requests.

Additionally, the Le Chat assistant, based on Medium 3.5, now supports a Work Mode for multi-step tasks such as email parsing, calendar checks, and Jira ticket creation.

Source: mistral.ai