Mozilla and Mistral announced their Firefox partnership on September 16, 2026, bringing Mistral models to Firefox Smart Window beta. Mistral Small 4 is now a model option, and Smart Window is rolling out in France with French-language support—the feature’s first expansion beyond North America. In the United States, it remains part of the North American beta rollout.

What changes in Firefox Smart Window

Smart Window is an optional beta browsing environment inside Firefox. It works separately from Classic and Private windows and can use selected browser context—such as open tabs, selected browsing history, chats and user-controlled memories—to help with research, summaries, comparisons and planning.

It also handles browser housekeeping. Smart Window can group related tabs, identify duplicate tabs and search browsing history for pages you viewed previously. That makes it less like a chatbot sitting beside the browser and more like an assistant wired into the browser’s working context—provided you choose to give it that context.

The newly announced model option is Mistral Small 4. Mozilla and Mistral presented the integration as a way to bring more provider choice to browser-based AI rather than tying the feature to a single model family.

An AI window inside Firefox

Smart Window is designed for tasks that involve several pages at once: comparing information across tabs, summarizing what you have open, organizing research and finding an older page in your browsing history. The beta label matters here. This is an optional environment still being developed, not a replacement for Firefox’s Classic or Private windows.

The feature is also opt-in. Mozilla says users decide whether to enable and use Smart Window, and can manage it through Firefox’s AI Controls.

Model choice and user control

Earlier Firefox Nightly Smart Window settings showed selectable model backends and custom model endpoint fields.

Mozilla says users can choose among model options and configure a locally run model. An earlier Firefox Nightly build showed how that approach worked in practice: its settings exposed fields for a model name, model endpoint and API key or authentication token.

That earlier configuration, demonstrated on February 9, 2026, listed three different backends: Gemini flash-lite for Fast, Qwen2-235B A22B-throughput for Flexible, and gpt-om-120b for Personal. Those labels belonged to that earlier Nightly build; the September partnership separately identifies Mistral Small 4 as the new Smart Window option.

The model-choice idea is important because it gives Smart Window a different shape from an assistant locked to one provider. You can use the built-in options Mozilla exposes, and Mozilla says you can also bring a locally run model into the configuration.

Privacy: what Mozilla says

Mozilla says Smart Window requests pass through Mozilla first, so a selected model provider receives Mozilla’s IP address rather than the user’s IP address. Mozilla also says it does not retain chat transcripts, use conversations to train models or sell user data. Memories, the company says, are stored on the user’s device.

Those controls are part of Mozilla’s stated design for the feature. They sit alongside the opt-in switch and model selection, giving users several points of control over whether Smart Window runs and what model handles their requests.

Where Smart Window is rolling out

The partnership announcement identifies North America and France for the supported rollout. France is the first European expansion and includes French-language support. Mistral said the United Kingdom and Germany were expected to follow later in 2026.

For US users, the feature remains within the North American beta. Smart Window is a Firefox feature rather than a separately priced product, and no subscription price is attached to the Mistral integration.

Mozilla’s browser AI experiment is therefore expanding in two directions at once: geographically, with France as its first European market, and architecturally, by putting Mistral Small 4 alongside a broader promise of model choice.