In this talk, I will share updates on Chrome Built-in AI in 2025, such as the following:
- Gemini Nano model is enhanced based on Gemma 3n with more tokens and a larger top-K
- the latest status of the APIs
- Multimodal inputs in Promp API, such as audio and image
- WIP Proofreader API for fixing errors in English input
- Translation, Language Detection, and Summarizer APIs are stable in Chrome 138
- Firebase experimental hybrid on-device inference that chooses either on-device or cloud models based on their availability.