Gemini 2.5 Pro (Preview 03-25): benchmark results
March 25 Gemini 2.5 Pro preview snapshot. Provider: Google. Released 2025-03-25. Access: API.
Aggregate Rating 1746 ± 7, rank #123 of 1656 rated models, from 1365 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BALROG BabaIsAI (VLM) | 53.33 | Progress (%) | 100 |
| BALROG Crafter (VLM) | 37.27 | Progress (%) | 100 |
| BALROG NetHack (VLM) | 1.64 | Progress (%) | 100 |
| BALROG Official VLM Leaderboard | 35.7 | % Progress (percentage) | 100 |
| BuzzBench (Humour) | 71.09 | Humour Score (0-100) | 100 |
| CodeGolf Bench | 73.9 | C++ Best %ile (self-reported) | 100 |
| CodeGolf Bench - Python Best Percentile | 68.04 | Best percentile (0-100): for each problem, the best of the s | 100 |
| EuroEval Faroese NLU - FoSent | 91.61 | Matthews correlation coefficient (%) | 100 |
| ForecastBench Baseline - Zero-Shot - Dataset | 61.4 | Dataset Brier Index (difficulty-adjusted, higher is better) | 100 |
| HELM VHELM - A Okvqa Base | 93 | PEM | 100 |
| HELM VHELM - A Okvqa Chinese | 89.3 | PEM (Translate) | 100 |
| HELM VHELM - A Okvqa Dialect | 91.9 | PEM (Fairness) | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-pro-preview-03-25 · How It Works · Data refreshed daily, snapshot 2026-10-20.