Google DeepMind

Unsloth has released GGUF quantizations of EmbeddingGemma 2, Google DeepMind's open 740M-parameter embedding model for text, images, video and audio. The base model maps all four modalities into one 768-dimensional space with an 8K context window under Apache 2.0.

Google-affiliated researchers Benjamin Bratton, Blaise Agüera y Arcas and James Manyika propose "Artificial Symbiotic Intelligence," a framework in which AGI emerges from a social system of people and AI agents rather than a single self-improving machine. The essay, written for the Deepmind Institute, is a conceptual argument and reports no benchmark results.

Google DeepMind has added Live Avatar, a real-time video avatar capability, to its Gemini 3.8 Live conversational model. The feature is available now in Gemini Enterprise, supporting lip-synced, expressive avatars across 97 languages with background tool calling.

Google DeepMind's new head Koray Kavukcuoglu says Gemini 4 is in early post-training and could ship well before year-end, following the quiet cancellation of Gemini 3.5 Pro. He downplayed the AGI question that defined predecessor Demis Hassabis's tenure, calling it 'not the right conversation.'

Google and DeepMind researchers introduced Dream-RSI, a method that lets AI agents test new search strategies by replaying recorded past attempts instead of running costly new computations. Tested on Gemini 3.1 Pro and Gemini 3.7 Flash across eight tasks, it matched or beat baselines while using far fewer attempts.

Google DeepMind Institute researchers Rohin Shah and Anca Dragan argue that visible chain-of-thought reasoning is a key safety mechanism for catching deceptive AI behavior, but say OpenAI's GPT-6 Astra system card already shows a significant drop in how well that reasoning can be monitored.

Google DeepMind has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two voice-dialogue models that reason and execute background tasks without interrupting conversation. Google claims the Extended Thinking model ranks #1 on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6.

Google DeepMind has released AlphaGenome Atlas, a free research platform containing precomputed predictions for the effects of all 9 billion possible single-letter DNA variants in the human genome. The 1-petabyte dataset includes a new AlphaGenome Variant Impact (AVI) score that combines AlphaGenome and AlphaMissense predictions to help researchers rank genetic variants by likely impact.

Google DeepMind tasked 100 AI agents running on Gemini 3.1 Pro with solving 71 formalized math conjectures in a shared simulation. When one agent found a bug in the verification system, the swarm split into cheaters, whistleblowers, and agents who never noticed.

Google DeepMind has upgraded its Co-Scientist system from a hypothesis generator into a closed-loop research partner that plans experiments, controls lab equipment, and writes manuscripts. Built on Gemini 3 models, the system cut fabricated-result rates from 46% to 4% using new verification modules, according to Google.

Google DeepMind published a retrospective on 15 years of game-based AI research, tracing a line from 2015's Atari-playing DQN through AlphaGo, AlphaZero, MuZero, and AlphaStar to its current generalist agent, SIMA 2. The post also details a new research partnership with Fenris Creations, the studio behind EVE Online, to study continual learning, memory, and long-horizon planning in a persistent multiplayer universe.

Google DeepMind has released SL2T, a massively multilingual sign-language-to-text translation model trained on over 100,000 hours of data across 50+ sign languages. The model powers new sign-to-text dictation features in Gboard and Live Transcribe on Pixel 11, starting with American Sign Language to English.

Google DeepMind has released Gemini Robotics ER 2, an embodied reasoning model that acts as a high-level planning brain for robots, adding continuous video progress tracking, precision moment-finding, and multi-robot collaboration. The model is available now via the Gemini API and Google AI Studio, with private preview access on the Gemini Enterprise Agent Platform.

Google DeepMind researchers developed GenCeption, which repurposes Alibaba's Wan2.1 video generator for computer vision tasks including depth estimation, segmentation, and 3D pose estimation. The model matches state-of-the-art specialized systems while training on only 7,500 synthetic videos—between 7 and 500 times less data than competing approaches.

Google has released Gemini 3.1 Flash Lite Image, also called Nano Banana 2 Lite, which the company describes as its fastest and cheapest image generation model. The model is available through Google's AI Studio and Gemini API with the identifier gemini-3.1-flash-lite-image.

Google DeepMind released Nano Banana 2 Lite (gemini-3.1-flash-lite-image), its fastest image generation model with 4-second text-to-image latency priced at $0.034 per 1K-resolution image. The company also opened developer access to Gemini Omni Flash (gemini-omni-flash-preview) for video generation and editing at $0.10 per second of output.

NVIDIA released a quantized version of Google DeepMind's DiffusionGemma 26B A4B IT, a multimodal model with 25.2B total parameters (3.8B active) that processes text, image, and video inputs. The NVFP4-quantized model achieves generation speeds exceeding 1,100 tokens per second on NVIDIA H100 GPUs while supporting a 256K token context window.

Amazon Web Services today announced availability of Google DeepMind's Gemma 4 family on Amazon Bedrock. The open-weight models include three instruction-tuned variants spanning 2.3B to 30.7B parameters, with 256K context windows, multimodal input support, and built-in reasoning mode.

Google DeepMind released DiffusionGemma, a 26B parameter mixture-of-experts model that generates text using discrete diffusion instead of autoregression. The model processes blocks of 256 tokens in parallel, achieving generation speeds exceeding 1100 tokens per second on H100 GPUs in low-batch settings.

Google DeepMind has released Gemma 4 12B, a 12-billion parameter multimodal model that runs locally on laptops with 16GB of RAM. The model eliminates separate vision and audio encoders, processing raw inputs directly through its language model backbone under an Apache 2.0 license.