Mistral Moves Code Agents to the Cloud with Mistral Medium 3.5 Model

Mistral AI has announced the release of Mistral Medium 3.5, a new flagship model with 128 billion parameters (dense architecture) and a context window of 256 thousand tokens, which combines instruction following, reasoning, and coding within a single set of weights. The model is openly available under a modified MIT license, can run independently on four GPUs, achieves 77.6% on SWE-Bench Verified and 91.4 on τ³-Telecom. API pricing: $1.5 per million input tokens and $7.5 per million output tokens. This model powers remote agents in Mistral Vibe and Le Chat: coding sessions are executed in the cloud asynchronously and in parallel; users can launch them from the command line or web interface, track progress, and receive notifications upon completion (for example, as a pull request on GitHub). Also introduced is Work mode in Le Chat (preview), a powerful agent for multi-step tasks such as research, analysis, and cross-tool scenarios, with built-in connectors to email, calendar, documents, and messaging apps. All agent actions are displayed, and sensitive operations (sending messages, writing documents) require explicit user approval. The new capabilities are available on Pro, Team, and Enterprise plans.