Mistral ships Medium 3.5, its first merged flagship model at 128 billion parameters, and adds cloud coding agents and Work Mode to Le Chat

Mistral AI, the Paris-based large language model company founded in 2023, announced on April 29 a set of updates anchored by the release of Mistral Medium 3.5, a 128-billion-parameter open-weight model that the company describes as its first fully merged flagship. The model combines instruction-following, reasoning, and coding in a single set of weights, running with a 256,000-token context window and capable of self-hosting on as few as four GPUs. On SWE-Bench Verified, the benchmark for real-world software engineering tasks, Medium 3.5 scores 77.6%, ahead of Mistral's own Devstral 2 and models including Qwen3.5 397B A17B.

Two accompanying product launches are at least as consequential as the model itself. The first is Vibe remote agents, which shift Mistral's developer coding tool from local terminal execution to asynchronous cloud sessions. Where the original Vibe CLI required a developer to keep a terminal session active while the AI worked through code changes, remote agents run in isolated cloud sandboxes independent of any local process. A developer can kick off a coding task, close the laptop, and return to completed results. Integrations with GitHub, Jira, Linear, and Sentry allow agents to read issues, create branches, and file pull requests without manual handoffs. Local sessions can also be transferred to the cloud mid-run, preserving full session history and state.

The second addition is Work Mode for Le Chat, Mistral's consumer-facing assistant. Work Mode is an agentic interface for complex tasks requiring sequential tool calls: research synthesis, inbox triage, calendar management, structured searches across connected services. The agent requires explicit user approval before taking sensitive actions such as sending messages or modifying files. Both remote agents and Work Mode are available on Le Chat Pro ($14.99 per month), Team ($24.99 per seat per month), and Enterprise plans.

The model is available via API at $1.50 per million input tokens and $7.50 per million output tokens. Open weights are released under a modified MIT license, which allows commercial and non-commercial use but carves out restrictions for organizations above a high-revenue threshold. The shift away from Apache 2.0, which Mistral has used for previous open releases, marks a change in how the company balances openness with commercial sustainability. Companies large enough to benefit most from inference-scale deployment of the model will need a commercial arrangement rather than the fully open license.

Mistral raised $830 million in debt financing in March 2026 to fund AI data center infrastructure, a move that underlined how capital-intensive the company's infrastructure ambitions have become alongside its model work. Medium 3.5 arrives at a moment when the competitive field for developer AI tools is expanding rapidly, with GitHub Copilot, Cursor, and others competing for engineering workflows. Mistral's bet is that a model combining multimodal reasoning, coding, and instruction-following in a single open-weight deployment, available at competitive API pricing, gives enterprise buyers and developers a credible alternative that does not require routing sensitive data through American cloud providers.

The EU's AI Act high-risk compliance deadline, which remains in flux following the April 29 trilogue breakdown, will eventually shape how European organizations think about which models they can deploy in regulated contexts. Mistral's open-weight strategy positions it as the natural choice for organizations that need to audit what is running in their infrastructure. That positioning has always been part of the company's thesis. The question Medium 3.5 is trying to answer is whether the model is now capable enough that European enterprises choose it on merit rather than regulatory preference alone.

Sources

• Remote agents in Vibe. Powered by Mistral Medium 3.5. — Mistral, 2026-04-29
• Mistral AI Launches Remote Agents in Vibe and Mistral Medium 3.5 with 77.6% SWE-Bench Verified Score — MarkTechPost, 2026-05-02
• Mistral Launches Cloud Coding Agents with Medium 3.5 — AI Business Review, 2026-05-03
• Mistral's new flagship Medium 3.5 folds chat, reasoning, and code into one model — The Decoder, 2026-05-02