Gemini 3 Deep Think upgrade (Feb 2026)
Upgraded Deep Think, built on Gemini 3.1 Pro, reports 84.6% on ARC-AGI-2 (ARC Prize verified), 48.4% on Humanity's Last Exam and a 3455 Codeforces Elo.
Retuned with scientists and engineers for messy, real-world research problems rather than only contest math. Available to Google AI Ultra subscribers in the Gemini app, with early API access for researchers and enterprises.
- Date
- Thursday, 12 February 2026
- Lab
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| ARC-AGI-2 | 84.6% verified by ARC Prize Foundation | ARC Prize |
| Humanity's Last Exam (no tools) | 48.4% company-reported | company |
| Codeforces | 3455 Elo company-reported | company |
The model page calls this "Gemini 3.1 Deep Think (Feb 2026)".
Sources
- deepmind.google/blog/gemini-3-deep-think-advancing-science-research-and-engineering/
- deepmind.google/models/gemini/deep-think/
- blog.google/innovation-and-ai/products/google-ai-updates-february-2026/
This record was checked against its sources on 6 October 2026. How we check