Mistral has introduced Mistral Medium 3.5, a 128B dense model that combines instruction following, reasoning, and coding into a single set of weights. The model is open‑weight under a modified MIT license and can run on as few as four GPUs, making it suitable for self‑hosting. It powers new remote coding agents in Mistral Vibe, which run in the cloud and can be started from the Mistral Vibe CLI or directly in Le Chat. These agents can run long, parallel sessions, notify users when tasks finish, and can be teleported from local CLI sessions to the cloud. In addition, a new Work mode in Le Chat, also powered by Mistral Medium 3.5, provides a powerful agent for complex, multi‑step tasks such as research, analysis, and cross‑tool actions. The agent can call tools in parallel, produce structured output, and interact with systems like GitHub, Linear, Jira, Sentry, Slack, and Teams. The model scores 77.6% on SWE‑Bench Verified and 91.4 on τ³‑Telecom, and is available on Mistral’s Pro, Team and Enterprise plans. Pricing for the API is $1.5 per million input tokens and $7.5 per million output tokens. The release expands Mistral’s agentic capabilities and offers developers a way to offload coding tasks to the cloud while maintaining visibility and control over the process.
Remote agents in Vibe. Powered by Mistral Medium 3.5.
Mistral has released Mistral Medium 3.5, a 128B dense model that merges instruction following, reasoning, and coding. It powers new remote coding agents in Mistral Vibe and a Work mode in Le Chat that can run long, multi‑step tasks in the cloud. The model is open‑weight, runs on as few as four GPUs, and is available on Mistral’s Pro, Team and Enterprise plans.
Why it matters
The ability to run autonomous, long‑horizon coding agents in the cloud reduces developer bottlenecks and enables parallel task execution, a key step toward scalable AI‑assisted software development.