Gemini 3 Flash (Preview) (Minimal): benchmark results
Gemini 3 Flash (Preview) evaluated at the minimal reasoning-effort setting. Provider: Google. Released 2025-12-17. Access: API.
Aggregate Rating 1706 ± 17, rank #342 of 2168 rated models, from 39 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Alexandria - Human Dialectness Rating (Egyptian Arabic) | 3.62 | Dialectness and fluency (1-5; native Egyptian speakers rate | 100 |
| Alexandria - Human Dialectness Rating (Mauritanian Arabic) | 3.04 | Dialectness and fluency (1-5; native Mauritanian speakers ra | 100 |
| CADEngBench - Assembly End-to-End (A2) | 15.8 | End-to-end pass rate (%; share of 120 assembly relation task | 100 |
| CADEngBench - Assembly Grounding (Typed@1) | 23.3 | Typed top-1 accuracy (%; share of 120 assembly relation task | 100 |
| CADEngBench - Engineering Requirements (L1) | 30 | Requirement pass rate (%; share of 300 parametric part tasks | 100 |
| CADEngBench - Functional Editing (L2-E) | 72.3 | Edit pass rate (%; share of 300 functional-edit tasks on sup | 100 |
| GPTNT - Manual VQA Reading | 81.2 | Accuracy (%; 191 questions reading specific text or values f | 100 |
| GPTNT - Simulator VQA | 70 | Accuracy (%; 640 questions about rendered game scenes: readi | 100 |
| MultiGlobeQA (Parametric) | 22.7 | Exact-match accuracy (%; Tier 1: the question alone, answere | 100 |
| UGI - Natural Intelligence | 70.41 | NatInt Score | 96.4 |
| UGI - Writing | 65.05 | Writing Score | 94.6 |
| M-GATE - Grammatical Error Detection | 0.18 | Mean MCC over 30 languages (-1 to 1; temperature 0) | 90.7 |
Interactive version: theaggregate.ai/model?slug=gemini-3-flash-preview-minimal · How It Works · Data refreshed daily, snapshot 2026-10-20.