| Benchmark | Gemini 3.1 Pro | GPT-4o |
|---|---|---|
| AA Agentic Index | 23 | 9.7 |
| AA Intelligence | 47.7 | 11.2 |
| AA-LCR | 79 | 55 |
| AA-Omniscience | 32.9 | -10.5 |
| arc_agi_1 | 98 | 4.5 |
| ARC-AGI-2 | 77.1 | 0 |
| arena_vision | 1280 | 1162 |
| Artificial Analysis Coding Index | 68.8 | 24.2 |
| Chinese SimpleQA (C-SimpleQA) | 85.9 | 64.6 |
| Codeforces | 3052 | 759 |
| critpt | 17.7 | 0 |
| Fortress | 29.8 | 47.2 |
| gdpval | 23.3 | 0 |
| GPQA Diamond | 94.3 | 70.1 |
| HLE | 51.4 | 5.3 |
| IFBench | 77.1 | 36 |
| LiveCodeBench | 91.7 | 38.3 |
| LiveCodeBench v6 | 91.7 | 30.9 |
| mathvision | 89.8 | 30.4 |
| mmlu | 92.6 | 88.1 |
| MMLU-Pro | 91 | 74.7 |
| mmmlu | 92.6 | 81.4 |
| MMMU-Pro | 83 | 59.9 |
| multichallenge | 71.4 | 40.3 |
| MultiNRC | 64.7 | 12.4 |
| OmniScience Accuracy | 55.3 | 23.7 |
| OmniScience Non-Hallucination | 50.1 | 62.1 |
| OSWorld-Verified | 76.2 | 5 |
| scicode | 59 | 33.4 |
| simplebench | 79.6 | 17.8 |
| SWE-bench Verified | 80.6 | 38.8 |
| Terminal-Bench Hard | 68.5 | 8.3 |
| vectara_answer_rate | 99.4 | 93.8 |
| vectara_avg_summary_length | 107.7 | 86.6 |
| vectara_factual_consistency | 89.6 | 90.4 |
| vectara_hallucination_rate ↓ | 10.4 | 9.6 |
| τ²-Bench Telecom (AA run) | 99.3 | 28.9 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.