| Benchmark | Claude 4 Opus | Gemini 3 Pro |
|---|---|---|
| AA Intelligence | 31.7 | 40.6 |
| AA-LCR | 40 | 73 |
| AIME 2025 | 75.5 | 100 |
| arc_agi_1 | 27 | 75 |
| ARC-AGI-2 | 8.6 | 54 |
| arena_vision | 1207 | 1290 |
| Artificial Analysis Coding Index | 34 | 46.5 |
| Fortress | 27.6 | 41.7 |
| frontiermath_tier_4 | 4.2 | 18.8 |
| GPQA Diamond | 79.6 | 91.9 |
| HLE | 12.3 | 45.8 |
| hmmt_feb_2025 | 60 | 97.5 |
| IFBench | 53.7 | 70.4 |
| livebench | 74.6 | 73.4 |
| LiveCodeBench | 70.4 | 92 |
| LiveCodeBench v6 | 47.4 | 90.7 |
| longbench_v2 | 55.6 | 68.2 |
| metr_hcast | 61.7 | 71.3 |
| metr_re_bench | 20 | 35 |
| metr_swaa | 99.8 | 99.5 |
| MMLU-Pro | 86.6 | 90.1 |
| mmmlu | 88.8 | 91.8 |
| multichallenge | 58.6 | 65.7 |
| MultiNRC | 33.9 | 59 |
| scicode | 40.9 | 56.1 |
| SEAL VISTA | 47 | 51.5 |
| simplebench | 58.8 | 76.4 |
| simpleqa | 22.8 | 72.1 |
| swe_bench_bash | 67.6 | 74.2 |
| SWE-bench Verified | 79.4 | 78 |
| Terminal-Bench Hard | 31.1 | 56.9 |
| vectara_answer_rate | 91 | 99.4 |
| vectara_avg_summary_length | 123.2 | 101.9 |
| vectara_factual_consistency | 88 | 86.4 |
| vectara_hallucination_rate ↓ | 12 | 13.6 |
| τ²-Bench | 57 | 98 |
| τ²-Bench Telecom (AA run) | 73.4 | 98 |
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.