Gemini 2.5 Flash (Preview 04-17): benchmark results
April 17 Gemini 2.5 Flash preview snapshot. Provider: Google. Released 2025-04-17. Access: API.
Aggregate Rating 1684 ± 9, rank #237 of 1656 rated models, from 619 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AVI-Bench - Audio-Visual Captioning | 57.97 | Task score (%) on audio-visual captioning (280 videos): FENS | 100 |
| CodeGolf Bench - C++ Pass@1 | 43.24 | Pass@1 (%): share of problems whose C++ solution passes ever | 100 |
| ENEM Item Difficulty (Point-Based Prompt) | 0.37 | Spearman rank correlation (-1 to 1) between the model's 1-10 | 100 |
| HELM Safety - Xstest | 98.83 | LM Evaluated Safety score | 100 |
| KOFFVQA - Graph and Chart Understanding | 96.67 | Score (%) | 100 |
| Long-form RewardBench - Reasoning | 90.2 | Pairwise accuracy (%) on the math reasoning subset of Long-f | 100 |
| MEDIC Benchmark | 92.36 | MEDIC Public Table Average | 100 |
| SciVQR - Chemistry | 87.6 | Chemistry accuracy (%) with zero-shot chain-of-thought promp | 100 |
| SciVQR - Physics | 78.1 | Physics accuracy (%) with zero-shot chain-of-thought prompti | 100 |
| EuroEval Danish NLU - Angry Tweets | 63.76 | Matthews correlation coefficient (%) | 99.8 |
| EuroEval Faroese NLU - ScaLA FO | 51.83 | Matthews correlation coefficient (%) | 99.8 |
| EuroEval Faroese NLU - FoSent | 84.89 | Matthews correlation coefficient (%) | 99.7 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-flash-preview-04-17 · How It Works · Data refreshed daily, snapshot 2026-10-20.