Gemini 3 Flash (Preview) (Low): benchmark results
Provider: Google. Released 2025-12-17. Access: API.
Aggregate Rating 1640 ± 24, rank #556 of 2170 rated models, from 26 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CALRK-Bench - Judgment Shift Attribution | 70.7 | Accuracy (times 100) on the 100 Type-JSA questions of CALRK- | 100 |
| CALRK-Bench - Temporal Context Recognition | 77 | Accuracy (times 100) on the 100 Type-TCR questions of CALRK- | 100 |
| Context Arena MRCR (2-needle) | 89 | AUC@1M (%) | 97.5 |
| Context Arena MRCR (4-needle) | 63 | AUC@1M (%) | 92.7 |
| Context Arena MRCR (8-needle) | 40.9 | AUC@1M (%) | 88.6 |
| ObviousBench | 98.61 | Answer pass³ (%) | 83.2 |
| CALRK-Bench - Information Sufficiency Recognition | 74.8 | Accuracy (times 100) on the 279 Type-ISR questions of CALRK- | 80 |
| DeepResearchBench | 49.8 | Average Score | 75 |
| ImmersedPrivacy - Social Context Rating Error (Audio as Text) | 1.24 | Mean absolute error (Tier 2: the model's 1-5 appropriateness | 72.7 |
| Epoch AI - Mystery Game Puzzles | 26 | Score | 68.4 |
| Chess Bench LLM | 650 | Lichess-Classical-calibrated Glicko-2 rating (native signed | 63.6 |
| EgoGapBench | 39.2 | Accuracy (%; zero-shot multiple choice on 1,000 single-image | 62.5 |
Interactive version: theaggregate.ai/model?slug=gemini-3-flash-preview-low · How It Works · Data refreshed daily, snapshot 2026-11-05.