Mistral AI Launches Remote Agents and Mistral Medium 3.5 Model

The AI company Mistral AI has shipped what amounts to a significant infrastructure upgrade for its coding agent ecosystem, moving sessions from local terminals to cloud-based execution while introducing a new flagship model to power the shift.

At the core of this announcement is Mistral Medium 3.5, a dense 128B parameter model with a 256k context window that now serves as the default in both Vibe (the coding agent platform) and Le Chat (the consumer assistant). The model scores 77.6% on SWE-Bench Verified, a benchmark that tests whether models can resolve real-world GitHub issues from popular open-source repositories.

That score places it ahead of Devstral 2 and models like Qwen3.5 397B A17B. For developers who have spent years watching benchmarks inflate without practical gains, this is one of the more reliable proxies for actual software engineering ability.

The bigger product change, however, is remote agents in Vibe. Until now, coding sessions ran locally, meaning the agent was tethered to your laptop and your terminal. You had to keep your machine awake, your terminal open, and your attention focused on every step. That constraint is gone.

According to the official announcement, coding sessions can now work through long tasks while you're away. Multiple agents can run in parallel, and you stop being the bottleneck on every step the agent takes. You can start cloud agents from the Mistral Vibe CLI or directly from Le Chat, offloading a coding task without leaving the conversation.

One particularly useful feature for developers already mid-session: ongoing local CLI sessions can be teleported up to the cloud when you want to leave them running. Session history, task state, and approvals carry across. So you don't lose your place — you just move the work off your machine (which is a relief if your laptop battery is dying).

Each coding session runs in an isolated sandbox, including broad edits and installs. When the work is done, the agent can open a pull request on GitHub and notify you, so you review the result instead of every keystroke that produced it. This is the kind of workflow shift that matters more than raw parameter counts.

On the integration side, Vibe plugs into GitHub for code and pull requests, Linear and Jira for issues, Sentry for incidents, and apps like Slack or Teams for reporting. The agent sits between the systems engineering teams already use, with humans in the loop wherever they're needed.

It fits the high-volume, well-defined work that takes a developer's time without taking their judgment: module refactors, test generation, dependency upgrades, CI investigations, as well as bug fixes. The physical reality here is that you're no longer staring at a terminal waiting for a refactoring job to complete while your coffee gets cold.

Mistral Medium 3.5 itself is described as the company's first flagship merged model, handling instruction-following, reasoning, and coding in a single set of weights. The model is multimodal, with a vision encoder trained from scratch to handle variable image sizes and aspect ratios. Most vision-language models reuse pretrained encoders like CLIP, so building this component from scratch suggests Mistral prioritized flexibility in how the model handles real-world image inputs.

Reasoning effort is now configurable per request, so the same model can answer a quick chat reply or work through a complex agentic run. This is important for developers integrating the model via API — you can dial down compute for simple lookups and dial it up for multi-step reasoning tasks, without switching models. API pricing is $1.50 per million input tokens and $7.50 per million output tokens.

Beyond the coding agent upgrades, Mistral is also shipping Work mode in Le Chat — a new agentic mode for more general, multi-step tasks. Work mode is powered by a new harness and Mistral Medium 3.5. The agent becomes the execution backend for the assistant itself, so Le Chat can read and write, use several tools at once, and work through multi-step projects until it completes what you've asked.

Practically, this means things like cross-tool workflows — catching up across email, messages, and calendar; preparing for a meeting with relevant context pulled from multiple sources; or triaging an inbox and creating Jira issues from team discussions. Sessions persist longer than a typical chat reply, so an agent can keep going across many turns, through trial-and-error, and through to completion.

In Work mode, connectors are on by default rather than chosen manually, which lets the agent reach into documents, mailboxes, calendars, and other systems for the rich context it needs to take correct action. This is a significant usability shift from typical chat assistants, where you manually select tools before each session.

Transparency is a built-in feature rather than an afterthought: every action the agent takes is visible — you see each tool call and the thinking rationale. Le Chat will ask for explicit approval — based on your permissions — before proceeding with sensitive tasks like sending a message, writing a document, or modifying data.

The model is available as open weights under a modified MIT license, with self-hosting possible on as few as four GPUs. It was built for long-horizon tasks, calling multiple tools reliably, and producing structured output that downstream code can consume.

Whether this actually changes developer workflows at scale remains the real question. The infrastructure is there, the benchmarks are respectable, and the pricing is competitive. But the difference between a tool that works in demos and one that survives in production is often measured in edge cases, not benchmark scores.

For now, the public preview is live. Developers can start coding sessions directly in Le Chat, so a task described in chat runs on the same remote runtime as the CLI and the web, and comes back later as a finished branch or a draft PR. The rest is up to how well the agents handle the messy reality of actual codebases.

Source: Mistral AI official announcement

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn