| Benchmark | Claude 4 Opus | Gemini 3.1 Pro |
|---|---|---|
| AA Intelligence | 31.7 | 47.7 |
| AA-LCR | 40 | 79 |
| arc_agi_1 | 27 | 98 |
| ARC-AGI-2 | 8.6 | 77.1 |
| arena_vision | 1207 | 1280 |
| Artificial Analysis Coding Index | 34 | 68.8 |
| Fortress | 27.6 | 29.8 |
| frontiermath_tier_4 | 4.2 | 16.7 |
| GPQA Diamond | 79.6 | 94.3 |
| HLE | 12.3 | 51.4 |
| IFBench | 53.7 | 77.1 |
| livebench | 74.6 | 79.9 |
| LiveCodeBench | 70.4 | 91.7 |
| LiveCodeBench v6 | 47.4 | 91.7 |
| mmlu | 92.9 | 92.6 |
| MMLU-Pro | 86.6 | 91 |
| mmmlu | 88.8 | 92.6 |
| multichallenge | 58.6 | 71.4 |
| MultiNRC | 33.9 | 64.7 |
| scicode | 40.9 | 59 |
| simplebench | 58.8 | 79.6 |
| SWE-bench Verified | 79.4 | 80.6 |
| Terminal-Bench Hard | 31.1 | 68.5 |
| vectara_answer_rate | 91 | 99.4 |
| vectara_avg_summary_length | 123.2 | 107.7 |
| vectara_factual_consistency | 88 | 89.6 |
| vectara_hallucination_rate ↓ | 12 | 10.4 |
| τ²-Bench Telecom (AA run) | 73.4 | 99.3 |
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.