Two years after Firefox first tucked someone else's chatbot into a sidebar, the same browser is now treating language models the way it once treated search engines: as interchangeable parts rather than a single house brand.
Here, Mozilla and Mistral said that Mistral Small 4 would join the roster inside Firefox Smart Window, the opt-in beta mode that tries to use open tabs and recent history to pick up a task users left half finished.
In the U.S. and Canada the model becomes another choice. In France, Smart Window itself arrives for the first time with official French-language support. The United Kingdom and Germany are listed as later stops in 2026.
This is not Firefox's first pass at putting a large language model next to a webpage.
Mistral was already sitting in Firefox before anyone called it a partnership.
When Mozilla first put third-party chatbots in a sidebar in 2024, Le Chat was one of five names on the list, next to Claude, ChatGPT, Gemini, and HuggingChat. That early hook was a pane onto Mistral’s own consumer site.
The announcement on 16 September 2026 is a different arrangement.
According to Mistral in a separate post, its Mistral model, Small 4, runs inside Firefox Smart Window as one of the engines that can read the tabs users allow it to see. The company is no longer only a bookmark in the chrome. It is part of how the beta browsing mode answers questions about the page in front of you.
Small 4 itself is also older than the Firefox deal.
Mistral released it on 16 March 2026 under Apache 2.0. It is a mixture-of-experts model with about 119 billion total parameters and roughly 6 to 6.5 billion active per token, 128 experts with four used at a time, a 256,000-token context window, and text-plus-image input.
Mistral describes it as a single model that can run in a fast instruct mode or a slower reasoning mode instead of shipping separate products for each job.
The open-weight licence is why both companies can talk about an open-source alternative rather than a closed API exclusive. In Firefox, though, the typical Smart Window user is not downloading those weights. Inference is handled by Mistral as a contracted backend, with Mozilla paying for the calls in the official presets.
The bring-your-own-endpoint setting is the route for people who want the same family of models on their own machine.
The privacy terms attached to Small 4 in Smart Window are stricter on paper than the old sidebar hook.
Mozilla says Smart Window chats are not stored on its servers by default. Mistral agreed to zero data retention for the traffic it handles. Firefox staff have described the inference contracts for these third-party presets as processing the prompt in memory without keeping it for training, advertising, or human review.
That is a contract about the Smart Window path.
The sidebar chatbot is a different pipe. If users still open Mistral Vibe or Le Chat from the side of the window, they're using Mistral's consumer product and Mistral's ordinary account and retention rules.
Mixing the two features is easy in conversation and misleading in practice. One is a hosted chat site framed in Firefox. The other is a model call inside Mozilla’s assistant, with a retention clause written for that call.
None of this makes Mistral the default brain of Firefox.
Gemini and Qwen remain in the same picker. Local endpoints remain available.
AI Controls can disable Smart Window and the rest of the generative features from one screen, a change some users treated as the only part of the announcement that mattered.
Firefox still does not ship a flagship consumer model of its own for this surface. What changed in September is that the French lab which first appeared in Firefox as a sidebar destination now supplies one of the models that can sit behind a browsing mode designed around tabs and history.
The earlier experiment asked which chatbot you wanted parked beside the page. The current one asks which model should interpret the page, and Mistral Small 4 is the new answer that can be given in French as well as in English.
