Google DeepMind says a Sierra Leone classroom trial shifted Gemini use toward learning behavior: queries about how to tackle problems rose from 68% to 90%. The eight-week RCT covered 1,763 students across 12 schools.
Google DeepMind’s new audio model translates speech across more than 70 languages while preserving tone, pace, and pitch. The rollout spans Google Translate, Google AI Studio, the Gemini Live API, and Google Meet previews.
AI media provenance is moving into search and browsers, not just model demos. Google DeepMind says SynthID has watermarked over 100 billion items, while OpenAI, ElevenLabs, and Kakao will add the watermark to more generated content.
Google launched Gemini 3.5 Flash at I/O 2026 on May 19, making it generally available the same day. It outperforms Gemini 3.1 Pro on coding and agentic benchmarks while running 4x faster at 40% lower cost.
Google DeepMind has unveiled an experimental project to redesign the mouse cursor with AI — the first major rethinking of the pointer in over 50 years. Powered by Gemini, the AI-enabled pointer understands screen context and acts on voice and gesture commands.
Google DeepMind announced a research partnership with CCP Games, the developer of EVE Online, to use the game's complex player-driven universe as a sandbox for advancing AI research in memory, continual learning, and long-term planning.
Google DeepMind introduced the AI co-clinician, a multimodal AI agent research initiative to support healthcare workers. In blind evaluations, the system achieved zero critical errors in 97 of 98 primary care queries and matched physician-level performance on medication knowledge benchmarks.
Google DeepMind is inviting developers to showcase their best creations built with GeminiApp or Google AI Studio for the Google I/O 2026 main stage, highlighting protein simulators, physics engines, and math-based art.
The important medical AI story here is not replacement but reliability. Google DeepMind says its AI co-clinician produced zero critical errors in 97 of 98 realistic primary-care queries, while physicians still beat it overall in multimodal telemedicine simulations.
Google DeepMind is tying frontier models directly to a national research agenda. On April 27, 2026, it said Korea will get a new AI Campus in Seoul plus joint work with SNU, KAIST, and three AI Bio Innovation Hubs.
This turns Google DeepMind’s science stack into a named national partnership instead of another generic AI pledge. The plan starts with a Seoul AI Campus, work with SNU and KAIST, and an AlphaFold base already used by more than 85,000 researchers in Korea.
Google DeepMind’s new training stack matters because datacenter boundaries are turning into frontier bottlenecks. Decoupled DiLoCo trained a 12B Gemma model across four U.S. regions on 2-5 Gbps links, more than 20x faster than conventional synchronization while holding 64.1% average accuracy versus a 64.4% baseline.