Google has unveiled SignGemma, a groundbreaking artificial intelligence (AI) model set to revolutionize communication for those who use sign language. This innovative model, part of the Gemma series, is currently in testing phases at the company’s headquarters in Mountain View, with plans for its public debut later this year. In line with other models in the Gemma family, SignGemma will be available as an open-source tool, making it accessible to both individuals and businesses. The model was demonstrated during the Google I/O 2025 keynote and aims to enhance communication for individuals with speech and hearing impairments when interacting with those unfamiliar with sign language.
SignGemma Can Track Hand Movements and Facial Expressions
In a recent post on X (formerly Twitter), Google DeepMind highlighted a demonstration of the AI model along with its anticipated release date. This marks the second unveiling of SignGemma, following a brief introduction by Gus Martin, the Gemma Product Manager, at the Google I/O event.
We’re thrilled to announce SignGemma, our most capable model for translating sign language into spoken text. 🧏
This open model is coming to the Gemma model family later this year, opening up new possibilities for inclusive tech.
Share your feedback and interest in early… pic.twitter.com/NhL9G5Y8tA
— Google DeepMind (@GoogleDeepMind) May 27, 2025
During the presentation, Martin emphasized that SignGemma offers real-time text translation of sign language, facilitating smoother face-to-face interactions. The AI has been trained on various sign language styles but shows the highest accuracy when translating American Sign Language (ASL) to English.
As reported by MultiLingual, being an open-source model allows SignGemma to operate offline, making it ideal for deployment in regions with limited internet connectivity. The model is constructed on the Gemini Nano framework and employs a vision transformer to analyze and interpret hand movements, shapes, and facial gestures. This openness invites developers to create applications using the model while also offering Google opportunities for integration into its existing AI tools, such as Gemini Live.
Described by DeepMind as “our most capable model for translating sign language into spoken text,” SignGemma is poised for an official release in the coming months. Currently undergoing initial testing, Google has also set up an interest form to encourage individuals to express their desire to try the model and share feedback.
