Google DeepMind announces Gemini Ultra 3, claiming significant progress toward artificial general intelligence with enhanced reasoning capabilities and multimodal understanding.
Google DeepMind has unveiled Gemini Ultra 3, claiming the most significant advancement toward artificial general intelligence yet achieved. The new model demonstrates unprecedented reasoning capabilities across text, code, images, and video.
The breakthrough comes from a novel architecture that combines enhanced chain-of-thought processing with real-time information retrieval. Gemini Ultra 3 can now maintain context across much longer conversations and handle complex multi-step problems with significantly improved accuracy.
"We've reached a point where Gemini Ultra 3 can genuinely understand and reason about complex topics in ways that feel remarkably human," said Demis Hassabis, CEO of Google DeepMind. "This isn't just bigger models—it's fundamentally better reasoning."
The model achieved record-breaking scores on multiple benchmark tests, including a 94% accuracy rate on challenging graduate-level reasoning problems. Perhaps more importantly, the model shows improved calibration, meaning it better acknowledges when it doesn't know something rather than hallucinating answers.
Google plans to integrate Gemini Ultra 3 across its product suite, from Search to Workspace to Android. Enterprise customers will get access through Google Cloud's Vertex AI platform, with pricing that reflects the model's enhanced capabilities.
