Gemini 3 Flash (High): benchmark results
Gemini 3 Flash evaluated at the high reasoning-effort setting. Provider: Google. Released 2025-12-17. Access: API.
Aggregate Rating 1773 ± 11, rank #187 of 2168 rated models, from 124 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GR-Ben (Thinking) | 61.1 | Average F1 (%) over the nine GR-Ben subsets (four scientific | 100 |
| GR-Ben (Thinking) - Biology | 60.1 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Chemistry | 60.5 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Computer Science | 66.5 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Deductive Reasoning | 47 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Inductive Reasoning | 75.3 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Mixed-Form Reasoning | 68 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| GR-Ben (Thinking) - Physics | 62.7 | F1 (%) of first-error identification: the harmonic mean of t | 100 |
| InfiniteBM Liar's Dice | 1566.83 | Arena Elo (elo) | 100 |
| LexKairos (Thinking Mode) | 84.57 | Mean score (%; average of the nine sub-task scores, accuracy | 100 |
| LexKairos (Thinking Mode) - Bounded Event Retrieval | 88.72 | Accuracy (%; 594 items: retrieve the case events inside a ti | 100 |
| LexKairos (Thinking Mode) - Effective Date Inquiry | 85.77 | Accuracy (%; 485 items: generate the effective date of a nam | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-3-flash-high · How It Works · Data refreshed daily, snapshot 2026-10-20.