Mistral, the Paris-based open-weight lab, is building two new surfaces into the browser version of its Le Chat product: a dedicated Code section for in-browser coding and an Apps area that lets users build, host, and share small applications. TestingCatalog reported both features on June 18, noting they were spotted in development and not yet publicly available.
The Code section is positioned at the same navigation level as the existing Chat and Work tabs, a placement that signals intent rather than just convenience. Mistral’s coding tooling until now has lived inside a command-line interface, meaning the web product was falling behind what users could get by downloading a separate tool. Bringing coding into the browser closes that gap for developers who prefer a no-setup path, and it may also preview a future desktop client, though the source notes that connection remains speculative.
The Apps area is the more consequential addition. Users would be able to create applications that fetch data through connectors and execute multi-step logic, then share those apps with others. That description maps directly onto what Anthropic calls Artifacts and what OpenAI has built into ChatGPT as in-chat app generation. Mistral already ships a connector directory and a Workflows engine; an Apps surface would pull both into a single interface visible to non-technical users. The feature is marked as a work in progress, and TestingCatalog cautions that the real capability ceiling is hard to judge at this stage.
The timing ties into a broader product push. Mistral CEO Arthur Mensch confirmed in a June 16 post that a new model is coming this summer, described as large and sparse. That phrasing points to a mixture-of-experts architecture, where only a fraction of parameters activate per inference call, cutting cost while keeping headline parameter counts high. Mensch said it will ship as open weights, with July access opening first for research, government, and industry partners.
Putting the browser features and the summer model side by side, the strategic read is that Mistral is trying to grow from a model supplier into a full product platform. The open-weights commitment is its clearest differentiator: Mensch explicitly framed it as a trust argument, noting that a system you can only reach through someone else’s interface is one you cannot audit, inspect, or improve. That framing positions Mistral against OpenAI and Google on principle rather than just on benchmark scores, where the gaps remain significant.
For builders currently evaluating which frontier chat product to build on or recommend to their teams, the Apps feature is the one to watch. ChatGPT’s in-chat apps and Claude’s Artifacts are already in wide use; the question is whether Mistral can ship a comparable surface before the category consolidates around one or two dominant choices. A July partner preview of the new model will be the first indicator of whether this summer’s release lands close enough to frontier to make the platform bet credible.
Reported by TestingCatalog (Alexey Shabanov), published June 18, 2026.
