Gemini 2.5 Flash (Preview): benchmark results
Google preview Gemini 2.5 Flash model, an early Flash snapshot from the Gemini 2.5 family. Provider: Google. Released 2025-06-17. Access: API.
Unified ELO 1703 ± 34, rank #326 of 2656 rated models, from 12 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Tinybird AI SQL Benchmark - First-Attempt Success Rate | 98 | Questions answered with a valid query on the first attempt ( | 78.1 |
| Global-MMLU-Lite | 90.92 | Score (%) | 67.6 |
| Reward Hacking Benchmark (RHB) | 0.8 | Exploit rate (%) (self-reported) | 66.7 |
| EQ-Bench 3 | 1179.5 | Elo | 42.5 |
| EQ-Bench | 1164.5 | Normalized Elo (self-reported) | 41 |
| EQ-Bench Creative Writing v3 | 1105.2 | Elo | 39.4 |
| Tinybird AI SQL Benchmark - Success Rate | 98 | Questions answered with a valid query within 3 retries (%) | 35.5 |
| Creative Writing v3 | 1221.8 | Elo score (self-reported) | 31.8 |
| ARC-AGI-1 | 33.33 | Accuracy (%) | 28.9 |
| ARC-AGI-2 | 1.69 | Accuracy (%) | 22.9 |
| Tinybird AI SQL Benchmark - Exactness | 39.9 | Result exactness vs human reference queries (0-100) | 20.2 |
| SciArena | 932.8 | Elo Rating | 18.9 |
Interactive version: theaggregate.ai/model?slug=gemini-2-5-flash-preview · How It Works · Data refreshed daily, snapshot 2026-09-19.