We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
DeepMind Blog·Feb 12·5 min read
Gemini 3 Deep Think: Advancing science, research and engineering
TL;DR
Researchers and engineers face challenges in solving complex scientific problems due to limitations in existing tools. Gemini 3 Deep Think, an upgraded reasoning mode, has been developed to address these challenges with enhanced mathematical and algorithmic capabilities.
✦ Why It Matters
Engineers and researchers can leverage Gemini 3 Deep Think to enhance their problem-solving capabilities in complex scientific tasks.
Key Takeaways
- Access Gemini 3 Deep Think through the Gemini app if you're a Google AI Ultra subscriber.
- Utilize the Gemini API for early access to test Deep Think in your projects.
- Leverage Deep Think to identify flaws in complex mathematical and scientific research.
- Apply Deep Think for optimizing engineering processes, such as material fabrication.
Full Summary
The updated Gemini 3 Deep Think is a specialized AI model designed to address intricate problems in science, research, and engineering. Developed in collaboration with experts, it excels in areas where data is often incomplete or messy, providing practical solutions rather than just theoretical insights.
Early testers have reported significant successes, such as identifying logical flaws in complex mathematical papers and optimizing fabrication methods for semiconductor materials. The model has achieved impressive benchmarks, including a score of 48.4% on Humanity’s Last Exam and 84.6% on ARC-AGI-2.
Additionally, it demonstrates gold medal-level performance in the International Physics and Chemistry Olympiads. This upgrade aims to empower researchers and engineers to interpret complex data and model physical systems effectively.