Mistral AI introduced remote agents for cloud-based coding through its Vibe platform, powered by Mistral Medium 3.5, on May 22, 2026. These agents allow users to run coding tasks asynchronously in the cloud, either from the Vibe CLI or directly in Le Chat, without leaving the conversation. The new Work mode in Le Chat extends this capability with an agent for complex, multi-step tasks like research and analysis. Mistral Medium 3.5, a 128B dense model with a 256k context window, is now the default model in Vibe and Le Chat, replacing Devstral 2 in coding agents. It is available in public preview and runs efficiently on as few as four GPUs. The model scores 77.6% on SWE-Bench Verified, outperforming Devstral 2 and models like Qwen3.5 397B A17B. It also achieves 91.4 on τ³-Telecom, demonstrating strong agentic capabilities. Remote agents can run in parallel, with sessions persisting across multiple turns. Users can inspect agent progress, including file diffs and tool calls, while the agent handles tasks like module refactors and bug fixes. Vibe integrates with tools like GitHub, Linear, Jira, and Slack, allowing agents to open pull requests and notify users upon completion. Coding sessions run in isolated sandboxes and can be teleported to the cloud from local CLI sessions. Mistral Medium 3.5 is accessible via API at $1.5 per million input tokens and $7.5 per million output tokens, with open weights available on Hugging Face under a modified MIT license. *Source: [mistral](https://mistral.ai/news/remote-agents-in-vibe-powered-by-mistral-medium-35/)*