Google DeepMind has announced Gemini Ultra 2, the next generation of its most capable AI model, featuring real-time video understanding that can analyze and respond to live video streams with human-level comprehension.
Breakthrough Capabilities
Gemini Ultra 2 processes video at 60 frames per second while maintaining context across hours of footage, enabling applications previously impossible for AI systems.
- Real-time sports analysis with tactical overlay generation
- Medical procedure assistance through live surgical video understanding
- Industrial quality control via continuous manufacturing line monitoring
- Accessibility features including real-time scene description for visually impaired users
Benchmark Results
On the new VideoQA-2026 benchmark, Gemini Ultra 2 scored 89.3%, surpassing human evaluators' average of 84.1% on complex video comprehension tasks. The model is available through Google Cloud Vertex AI with enterprise pricing starting at $0.005 per second of video processed.
