Mistral Launches Medium 3.5 Open-Weight Flagship and Remote Coding Agents in Vibe

Mistral AI

Models / LLM official + media 3 src. ~1 min

Mistral AI released Medium 3.5, a 128B dense open-weight model with a 256k context window scoring 77.6% on SWE-Bench Verified, released under a modified MIT license. Alongside the model, Mistral launched remote cloud-based coding agents in its Vibe platform that can run asynchronous long-running sessions and open pull requests on GitHub. A new Work mode for Le Chat powered by Medium 3.5 adds multi-step agentic workflows with cross-tool support.

Why it matters

Medium 3.5 is Mistral's first merged flagship model competitive with closed frontier coders as open weights, self-hostable on four GPUs. The Vibe remote agents put the European AI challenger directly into GitHub Copilot and Claude Code territory for agentic software development.

Importance: 4/5

Frontier-competitive open-weight coding model with SOTA SWE-Bench score paired with a full agentic coding platform launch.

Sources