With vibe coding growing in popularity, Mistral AI introduced remote coding agents in Vibe and a Work mode in its AI chatbot, all powered by a new model.
Mistral revealed on April 29 that its cloud coding agents can run for extended periods independently without supervision. The remote agents in Vibe are powered by Mistral Medium 3.5, the French startup’s new open-weight default model, as are the coding agents Mistral Vibe and Le Chat. The model is built to run for long stretches on coding and productivity, Mistral said. It has 128B parameters with a 256K context window, which allows it to manage instruction-following, reasoning and coding.
Mistral’s remote agents and model update exemplify the rapid growth of vibe coding in just the past year. The popularity of coding with AI models, using natural language, has exploded since OpenAI co-founder Andrej Karpathy coined the term in 2025. Mistral has moved heavily into it with the release of various models and tools that prioritize natural language over line-by-line coding. Last year, the Paris-based vendor introduced Mistral Vibe CLI, its command-line interface that developers can use to describe tasks in English. It also released Vibe 2.0 earlier this year, which enables agents to ask for clarification when a prompt is unclear.
A Unified Interface
With the new model, Mistral is seeking to make it easier for developers to do more with less, said William McKeon-White, an analyst at Forrester Research.
“This 'making it easier' also allows you to do more from whatever interface you want,” McKeon-White said. He added that with the model and remote-vibe coding, developers can access a coding terminal or interface directly connected to a code repository, either using the web or locally. This differs from Anthropic in that Claude Code has a separate interface, distinct from Claude itself.
“Mistral helps keep work more in context and allows for easier prompts to research to code workflows, while still allowing you to interact in a CLI,” McKeon-White said.
While enterprises need the connectivity that Mistral is bringing, Medium 3.5 is not necessarily better than other models on the market when it comes to coding or helping enterprises to create and deploy coding agents, McKeon-White continued.

"But you can easily run many more [coding agents] than most other competitors,” he said. He added that with the remote vibe coding agents, enterprise developers can leave them running for an extended period , enabling them to perform more tasks in parallel rather than in sequence.
However, how Mistral plans to keep its system running is still unclear, McKeon-White said.
“My big questions are around long-running memory and model context and how they plan to help their system keep track of what people are doing over multiple sessions,” he said.
New Agentic Mode
Mistral also rolled out a new agentic model, called Work Mode, in Le Chat. Introduced in preview, Work Mode is powered by a new harness and Mistral Medium 3.5. It enables Le Chat to read/write and use multiple tools at once. This means that users can catch up on emails, messages and calendars all in Le Chat through Work Mode. Users can also research topics across the web and then produce a brief or report that can be edited before sending, Mistral said.
While OpenAI provides a similar experience with its Agents SDK and connectors that can be connected to ChatGPT, “Mistral is trying to provide an alternative but something that is more enterprise-grade or enterprise-ready,” Su said.
He added that the AI chatbot has always been a source of mistrust for enterprises. Still, the Mistral approach of using a chat interface people are used to, with an orchestration layer that manages agents, models and routes them to the right task, is a "pretty compelling value proposition," he continued.
“The bundling of the services does broaden their appeal to both the consumer and enterprise markets, which is not necessarily something that their competitors have done,” Su said. He added that the vendor is well known in Europe and U.S., but still relatively unknown in the APAC market.