Gemini 3 Pro (Preview): benchmark results
Google's Gemini 3 Pro preview model for advanced reasoning and multimodal tasks. Provider: Google. Released 2025-11-18. Access: API.
Aggregate Rating 1799 ± 4, rank #66 of 1656 rated models, from 1926 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA LiveCodeBench | 91.75 | Pass@1 (%) | 100 |
| AA MMLU-Pro | 89.76 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Java | 55 | Accuracy (%) | 100 |
| AEGIS (Academic Image Forensics) - Textual Artifact Recognition | 84.67 | Macro-F1 (%) over two classes, telling whether the text regi | 100 |
| AI Chess Leaderboard (Continuation) | 1810 | Elo | 100 |
| AIDABench - Data Visualization | 78.97 | Score (%) on the 116 data-visualization tasks, best of three | 100 |
| AIDABench - Easy | 86.97 | Score (%) on the 178 easy tasks (at most 6 expert operation | 100 |
| AJ-Bench (LLM-as-a-Judge) | 75.05 | Overall F1 (%), mean of the seven subdomain F1 scores and av | 100 |
| AJ-Bench (LLM-as-a-Judge) - PPT | 76.1 | F1 (%) on PowerPoint GUI tasks (OSWorld; 21 tasks, 42 trajec | 100 |
| AJ-Bench (LLM-as-a-Judge) - Postgres | 73.2 | F1 (%) on Postgres data-system tasks (MCPMark; 18 tasks, 100 | 100 |
| AJ-Bench (LLM-as-a-Judge) - Wide Search | 72.7 | F1 (%) on Wide search (WideSearch-derived item checks; 9 tas | 100 |
| AJ-Bench (LLM-as-a-Judge) - Word | 72.14 | F1 (%) on Word GUI tasks (OSWorld; 12 tasks, 24 trajectories | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-3-pro-preview · How It Works · Data refreshed daily, snapshot 2026-10-20.