Mistral AI has introduced remote coding agents powered by its new Mistral Medium 3.5 model, now in public preview. The technology allows developers to run coding tasks in the cloud, enabling asynchronous execution while they step away from their work. Users can initiate these sessions from the Mistral Vibe CLI or directly in Le Chat, offloading tasks without leaving their conversation.
The model, a dense 128B parameter system with a 256k context window, is designed for long-running coding and productivity work. It is the default model in Mistral Vibe and Le Chat, replacing Devstral 2 in coding agents. The new Work mode in Le Chat extends this capability, allowing agents to handle complex, multi-step tasks like research, analysis, and cross-tool actions.
Sessions run in isolated sandboxes, with the ability to teleport local CLI sessions to the cloud. Developers can receive updates through file diffs, tool calls, and progress states. The model scores 77.6% on SWE-Bench Verified, outperforming Devstral 2 and other models like Qwen3.5.
It also excels in agentic capabilities, achieving 91.4 on τ³-Telecom. Mistral Medium 3.5 is available for use in Mistral Vibe and Le Chat, with API pricing at $1.5 per million input tokens and $7.5 per million output tokens. Open weights are accessible on Hugging Face under a modified MIT license.
Source: mistral
