Gemini 2.5 Flash: benchmark results
Google's speed-oriented Gemini 2.5 Flash model, balancing capability with lower latency. Provider: Google. Released 2025-06-17. Access: API.
Aggregate Rating 1680 ± 2, rank #245 of 1656 rated models, from 4054 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 360CityArena - Landmark Search (Language) | 28 | Success rate (%; the agent must stop within 10 m of a landma | 100 |
| 360CityArena - Localization | 12 | Exact-match accuracy (%; the agent infers its starting cell | 100 |
| 360CityArena - Object Count | 24 | Mean relative accuracy (%; count of specified objects in a d | 100 |
| AGC-Bench - ocw_connections | 1.19 | Dataset z-score | 100 |
| AICA-Bench - Emotion Understanding (Basic) | 68.05 | Weighted F1 (%) of emotion classification (expressed emotion | 100 |
| AICA-Bench - Emotion Understanding (CoT) | 69.32 | Weighted F1 (%) of emotion classification (expressed emotion | 100 |
| AICD Bench - Fine-Grained Human-Machine Classification (Zero-Shot CoT) | 28.14 | Macro-F1 (%; four-way human-written, machine-generated, hybr | 100 |
| AICD Bench - Fine-Grained Human-Machine Classification (Zero-Shot) | 25.95 | Macro-F1 (%; four-way human-written, machine-generated, hybr | 100 |
| AICD Bench - Robust Binary Classification (Zero-Shot CoT) | 62.31 | Macro-F1 (%; human-written vs machine-generated code under u | 100 |
| AMemGym - Overall Accuracy | 44.8 | Overall score (times 100): mean multiple-choice question-ans | 100 |
| Absurd World - Few-Shot - LEAST | 100 | Determine-Outcome Accuracy (%) | 100 |
| AcuityBench (Conversational) | 78 | Exact acuity match (%) in the conversational format on the 5 | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-flash · How It Works · Data refreshed daily, snapshot 2026-10-20.