Created:
11 hours ago
Smart Window lets you choose which AI model powers the built-in assistant. It offers 3 models to choose from and, for more advanced users, the option to bring your own model.
You can switch between models at any time to find the one that works best for you.
Table of Contents
- 1 Available models
- 2 Choose or change a model
- 3 Bring your own model
- 4 How models are chosen by the Firefox team
- 5 FAQ
- 5.1 Which model should I choose?
- 5.2 Can I change my model after my initial selection in onboarding?
- 5.3 How can I further customize my chosen model?
- 5.4 Does my conversation history go to the model providers?
- 5.5 Do different models handle my data differently?
- 5.6 Are some models safer or more secure than others?
- 6 Related articles
Available models
Smart Window offers several models with different strengths. While all models can perform most tasks, they may respond differently depending on your needs. The best way to see what model is best for you is to give each a try. Use the table below to get a sense for what you can expect from each model.
| Fast: Answers quickly | Flexible: Solid fit for most needs | Personalized: Most tailored answers | |
|---|---|---|---|
| Try it when… | You want quick and short answers | Your tasks vary from simple to complex | You prefer more personalized assistant responses |
| Model name | gemini-2.5-flash-lite | Qwen3-235B-A22B-Instruct-2507 | gpt-oss-120B |
| Model maker | Alibaba | OpenAI | |
| Model card (detailed specs) | Link | Link | Link |
Choose or change a model
You’ll be asked to pick a model the first time you use Smart Window, but you can switch models anytime, as often as you like, in Settings.
- Open .
- Go to .
- Select
- Make a choice in the section.
Bring your own model
Beyond the available choices, more advanced users can choose a custom option, to connect their own model. Go to the Custom Models article to learn more.
How models are chosen by the Firefox team
Firefox evaluates models before making them available in Smart Window. These evaluations focus on:
- How well models perform common tasks (for example, summarizing and answering questions)
- The language models offered in Smart Window are based on technical evaluations completed by the Firefox engineering team. While public benchmarks are available and are published by the model makers, the Firefox team takes the additional step of completing these internal evaluations to ensure that the models behave well in the context of Firefox and Smart Window use cases. For example: web page summarization, page Q&A, and more specific use cases performed by web browser users, such as product comparisons.
- How they handle sensitive or harmful content
- Additional Trust and Safety evaluations are performed to evaluate model handling of certain sensitive and harmful content. You can learn more about that here.
- How useful they are for real browsing scenarios
- Comparative evaluations are done to determine the promoted “characteristic” of the model. For example, gemini-2.5-flash-lite was seen to produce shorter results, with a higher token output rate, than the other models, hence why it has been labeled as the “Fast” model choice. This labeling is intended to help users who are unfamiliar with the individual models to make a choice for Smart Window.
- The usage limit
- The cost of the models to host and run is taken into consideration as well. Smart Window has a usage quota for each user to allow for generous but reasonable daily usage. When a user reaches their usage limit, they will be prevented for further use until the next day when the quota resets.
FAQ
Which model should I choose?
This depends on the experience you would like to have with Smart Window’s assistant. If you prefer fast responses with short answers, select the “Fast” option, which is gemini-2.5-flash-lite. While all models support personalization based on your browsing habits if Memories are enabled, if you like to have your responses personalized more frequently, select the Personalized option, which is gpt-oss-120B. If you are not sure, or don’t have a preference, select the Flexible option, which is Qwen3-235B-A22B-Instruct-2507.
Can I change my model after my initial selection in onboarding?
Yes. Your model can be changed at any time from AI Controls for Smart Window. See the instructions above for how to change your model.
How can I further customize my chosen model?
The three available models do not currently allow for further customization, outside the creation of individual memories. However, if you want to provide your own language model for the Smart Window assistant, you can follow the instructions above about how to set up your own custom model.
Does my conversation history go to the model providers?
No. The AI providers or model makers do not store your conversation. It may be briefly processed to generate a response, but it is not saved or kept after that.
Do different models handle my data differently?
No. All models available in Smart Window follow the same privacy and data handling standards set by Mozilla. Your data is consistently protected, regardless of which model you choose. To learn more, see Trust and Safety in Smart Window.
Are some models safer or more secure than others?
All models offered in Smart Window go through Mozilla’s Trust and Safety evaluations before being made available. These evaluations help ensure that models respond appropriately and handle sensitive content responsibly. To learn more, see Trust and Safety in Smart Window.
