Overview
- Updated Deep Think is live in the Gemini app for Google AI Ultra subscribers, with researchers and enterprises invited to request early API access.
- Google reports 48.4% on Humanity’s Last Exam without tools, 84.6% on ARC-AGI-2 verified by the ARC Prize Foundation, and a 3455 Elo on Codeforces.
- Performance claims extend to science benchmarks, including gold-medal level on the 2025 Physics and Chemistry Olympiads and 50.5% on the CMT-Benchmark for theoretical physics.
- Early users cited by Google include Rutgers’ Lisa Carbone finding a logical flaw in a math paper, Duke’s Wang Lab designing a thin-film growth recipe, and Google R&D lead Anupam Pathak accelerating component design.
- Google highlights practical tasks such as interpreting complex data, modeling physical systems in code, and converting a hand-drawn sketch into a 3D‑printable file.