Gemini 3.5 Flash (High): benchmark results
Gemini 3.5 Flash evaluated at the high reasoning-effort setting. Provider: Google. Released 2026-05-19. Access: API.
Aggregate Rating 1800 ± 8, rank #139 of 2168 rated models, from 309 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA APEX-Agents | 47.05 | Pass@1 (%) | 100 |
| AeroCopilotBench - Tier-2 Ineffective-Action Rate | 5.5 | Ineffective write attempts (%; lower is better) | 100 |
| AeroCopilotBench - Tier-2 Safety Compliance Rate | 99.5 | Episodes without a hard safety violation (%) | 100 |
| Computer Anthology Terminal Tasks (Gemini CLI) | 24.4 | pass@1 (%) | 100 |
| GroundBench | 57.7 | Filled-region IoU (0-100), IoU x 100 (mean over the five ver | 100 |
| GroundBench - 16 Vertices | 64.7 | Filled-region IoU (0-100), IoU x 100 (N=16 vertex budget; 1, | 100 |
| GroundBench - 24 Vertices | 65.9 | Filled-region IoU (0-100), IoU x 100 (N=24 vertex budget; 1, | 100 |
| GroundBench - 32 Vertices | 57.8 | Filled-region IoU (0-100), IoU x 100 (N=32 vertex budget; 1, | 100 |
| GroundBench - 8 Vertices | 62.6 | Filled-region IoU (0-100), IoU x 100 (N=8 vertex budget; 1,5 | 100 |
| LivingScreen | 69.3 | Success rate (%), mean of the three task tiers; LivingScreen | 100 |
| LivingScreen - Closed-Loop Application | 57.4 | Success rate (%) on L3 fact-checking, content moderation and | 100 |
| LivingScreen - GUI Action | 90.4 | Success rate (%) on L1 atomic GUI actions (liking, collectin | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-3-5-flash-high · How It Works · Data refreshed daily, snapshot 2026-10-20.