Gemini 2.5 Pro: benchmark results
Google Gemini 2.5 Pro model for advanced reasoning and multimodal tasks. Provider: Google. Released 2025-03-25. Access: API.
Aggregate Rating 1727 ± 2, rank #152 of 1656 rated models, from 4543 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 360Bench (CMP) | 46.5 | Overall accuracy (%) over all seven subtasks on 360Bench's 1 | 100 |
| 360Bench (CMP) - Direction Giving | 37.2 | Direction-giving accuracy (%) on multi-step route questions | 100 |
| 360Bench (CMP) - Fine-Grained Perception | 53 | Fine-grained perception accuracy (%), mean of instance recog | 100 |
| 360Bench (CMP) - Spatial Reasoning | 41.9 | Spatial reasoning accuracy (%), mean of object-to-object and | 100 |
| 360Bench (ERP) | 43.9 | Overall accuracy (%) over all seven subtasks on 360Bench's 1 | 100 |
| 360Bench (ERP) - Direction Giving | 44 | Direction-giving accuracy (%) on multi-step route questions | 100 |
| 360Bench (ERP) - Fine-Grained Perception | 53.3 | Fine-grained perception accuracy (%), mean of instance recog | 100 |
| 360Bench (ERP) - Spatial Reasoning | 40.5 | Spatial reasoning accuracy (%), mean of object-to-object and | 100 |
| AHA-Memes | 71.1 | Binary hateful versus not-hateful macro-F1 (%), zero-shot on | 100 |
| AHA-Memes - Fine-Grained Hate Types | 34 | Fine-grained macro-F1 (%) over ten hate and non-hate type la | 100 |
| AI for Education Visual Maths - Algebra | 100 | Accuracy (%) | 100 |
| AICA-Bench | 73.49 | Overall average on a percent scale: unweighted mean of the e | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-pro · How It Works · Data refreshed daily, snapshot 2026-10-20.