Overview
- Gemini 2.5 deep Think is now available in the Gemini app for Google AI Ultra members with a limited daily prompt allowance.
- The model leverages parallel thinking across multiple AI agents and extended inference time to explore and refine solution paths.
- Google credits early tester feedback and research breakthroughs for substantial benchmark gains since its May preview.
- Internal tests show Deep Think achieving 34.8% on Humanity’s Last Exam and 87.6% on LiveCodeBench V6, outpacing rival AI models.
- The gold-medal IMO variation will be shared with select mathematicians and academics as Google prepares API access for trusted developers and enterprises.