We recently released a feature in the Firefox mobile app called “Shake to Summarize”. The reception was remarkably positive, earning an honorable mention on Time Magazine’s best inventions of 2025.

Mozilla stands at the forefront of the open web movement, advocating for privacy, security, and accessibility online. Through its platform, Mozilla shares insights into the latest developments in web standards, browser technology, and internet policy. From  analyses of browser performance to explorations of emerging web technologies, Mozilla equips developers, designers, and internet users with the knowledge and tools they need to build a better, more inclusive web for all.

Mozilla

Mozilla's Firefox iOS team built 'Shake to Summarize', a feature that generates a webpage summary when the user shakes their phone. This post details the LLM model selection process, evaluating candidates across four dimensions: summary quality (using an LLM judge with GPT-4o), inference speed (time to first token and tokens/second), cost (Vertex AI pricing), and open-source alignment. Gemini 2.0 Flash led on quality, while Mistral Small proved the best overall fit — competitive quality, fastest speed, lowest cost at $0.10/M input tokens, and open weights. Mozilla ultimately chose Mistral Small to power the feature, released in September 2025.

Under the hood: The AI powering Firefox’s Shake to Summarize