Google DeepMind has released DiffusionGemma, a new approach that claims to accelerate text generation by a factor of four compared to conventional methods. The blog post announcement provides a headline-level overview without disclosing technical architecture details, benchmark data, or specific use cases. The name suggests the method applies diffusion-model techniques to the Gemma family of language models.
Google DeepMind announced Gemini 3.5 Live Translate, a feature that provides near real-time, natural voice translation. The capability is now available in Google AI Studio, Google Translate, and Google Meet. It delivers fluid, conversational translations, minimizing robotic outputs and reducing lag. This integration brings live speech translation directly into Google's widely used communication and development platforms.
Google DeepMind released Gemma 4 12B, a 12-billion-parameter open multimodal model. The model handles text and images without a separate vision encoder through a unified architecture. It is part of the Gemma family of open models. The announcement highlights the encoder-free design but provides no further performance or capability details.
A randomized controlled trial assessed the effectiveness of Gemini's Guided Learning feature. Results showed that the feature significantly boosted student engagement and accelerated learning outcomes. The study was conducted in Sierra Leone, with potential implications for education in other regions. This demonstrates the promise of AI-powered personalized learning tools.