Gemini 2.5 Pro
- Made by
- Released
- Announced in
- Gemini 2.5: Our most intelligent AI modelblog.google
- Event
- Gemini 2.5 Pro
- API record
- /api/v1/models/gemini-2-5-pro
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| Humanity's Last ExamNo tools · pass@1 | 18.8% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| reported later, in Start building with Gemini 3 | 21.6% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| SWE-bench Verifiedsingle attempt · reported later, in Start building with Gemini 3 | 59.6% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| multiple attempts: Google's scaffold draws several trajectories and the model picks one · parallel test-time compute · in Gemini 2.5: Our most intelligent AI model | 63.8% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| GPQA Diamondpass@1 | 84.0% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| reported later, in Start building with Gemini 3 | 86.4% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| ARC-AGI-2ARC Prize Verified · reported later, in Start building with Gemini 3 | 4.9% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| AIME 2024No tools · pass@1, averaged over multiple trials | 92.0% | CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.CorroboratedThe same number is printed in 1 other lab document: 1 |
| AIME 2025No tools · pass@1, averaged over multiple trials | 86.7% | CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.CorroboratedThe same number is printed in 1 other lab document: 1 |
| reported later, in Start building with Gemini 3 | 88.0% |
Sources
Reported by Google
- Gemini 2.5: Our most intelligent AI modelblog.google · 25 Mar 2025
Original image - Start building with Gemini 3blog.google · 18 Nov 2025
Original image
