Gemini 3 Pro: benchmark results
Google's Pro-tier Gemini 3 model for advanced reasoning and general tasks. Provider: Google. Released 2025-11-18. Access: API.
Aggregate Rating 1782 ± 3, rank #80 of 1656 rated models, from 1639 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 1D-Bench (Single-Round) | 79.6 | Final score (0-100) for single-pass generation of an executa | 100 |
| 1D-Bench (Single-Round) - Rendering Success Rate | 100 | Rendering success rate (%): share of instances whose single- | 100 |
| A2RBench | 40.9 | Total Acc (self-reported) | 100 |
| A2RBench - Symbolic Tasks (Remapped Symbols) | 34.8 | Accuracy (%) on the same 351 symbolic rule tasks after every | 100 |
| ADBench | 83 | Pass@3 (imported) | 100 |
| AGL1K | 3031.82 | Geo-score (0-5000) | 100 |
| AGL1K - Non-Speech Distance Error | 3727.33 | Mean distance error on clips without speech (km) | 100 |
| AGL1K - Speech Distance Error | 1548.29 | Mean distance error on clips with speech (km) | 100 |
| AISE-Bench | 56.06 | F1-LM (%) | 100 |
| ARFBench - Correlation | 71.2 | Accuracy (%) on the 170 Correlation questions (whether anoma | 100 |
| ATP-Bench | 79.88 | Final Score (0-100) assigned by the Chief Judge of a Gemini | 100 |
| ATP-Bench - Fashion | 79.83 | Final Score (0-100) assigned by the Chief Judge of a Gemini | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-3-pro · How It Works · Data refreshed daily, snapshot 2026-10-20.