Gemini 2.5 Pro (Preview 06-05): benchmark results
June 5 Gemini 2.5 Pro preview snapshot. Provider: Google. Released 2025-06-05. Access: API.
Aggregate Rating 1756 ± 11, rank #106 of 1656 rated models, from 126 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FactArena - Verdict Accuracy | 66.52 | Accuracy (%) | 100 |
| LLM Stats (Vibe-Eval) | 67.2 | Score (%) | 100 |
| M-JudgeBench | 64.76 | Overall pairwise accuracy (%), pooled over all ten subtasks, | 100 |
| M-JudgeBench - Long CoT Cross-Model Pairs | 72.52 | pairwise accuracy (%) on the 404 pairs where a correct and a | 100 |
| M-JudgeBench - Long CoT Same-Model Pairs | 70.03 | pairwise accuracy (%) on the 337 pairs where a correct and a | 100 |
| M-JudgeBench - Short CoT Cross-Model Pairs | 87.21 | pairwise accuracy (%) on the 297 pairs where a correct and a | 100 |
| M-JudgeBench - Short CoT Same-Model Pairs | 73.62 | pairwise accuracy (%) on the 326 pairs where a correct and a | 100 |
| M-JudgeBench - Visual Perception Error Detection | 99.19 | pairwise accuracy (%) on the 248 pairs where the rejected ch | 100 |
| SEAL - VISTA | 54.63 | Score | 98.4 |
| Kaggle MultiLoKo - Bengali | 42.8 | Exact Match (%) | 97.9 |
| Kaggle MultiLoKo - Korean | 52.8 | Exact Match (%) | 97.9 |
| Kaggle MultiLoKo - Thai | 39.6 | Exact Match (%) | 97.9 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-pro-preview-06-05 · How It Works · Data refreshed daily, snapshot 2026-10-20.