| Benchmark | Claude 4 Opus | Gemini 2.5 Flash |
|---|---|---|
| AA Intelligence | 31.7 | 17.7 |
| AA-LCR | 40 | 65.7 |
| aider_polyglot | 70.7 | 61.9 |
| AIME | 33.9 | 56.3 |
| AIME 2024 | 76 | 82.3 |
| AIME 2025 | 75.5 | 72 |
| AIR-Bench 2024 | 85.7 | 68.7 |
| anthropic_red_team | 98.9 | 98.8 |
| arc_agi_1 | 27 | 33.3 |
| ARC-AGI-2 | 8.6 | 2.5 |
| arena_vision | 1207 | 1214 |
| Artificial Analysis Coding Index | 34 | 22.2 |
| bbq | 99.3 | 97.7 |
| frontiermath_tier_4 | 4.2 | 4.2 |
| GPQA Diamond | 79.6 | 82.8 |
| harmbench | 91.7 | 62.6 |
| HLE | 12.3 | 12.1 |
| hmmt_feb_2025 | 60 | 64.2 |
| IFBench | 53.7 | 50.3 |
| LiveCodeBench | 70.4 | 76.2 |
| livecodebench_easy | 99.1 | 99.1 |
| livecodebench_hard | 33.1 | 48.3 |
| livecodebench_medium | 80.4 | 82.5 |
| MMMU | 73.7 | 79.7 |
| scicode | 40.9 | 35.9 |
| SEAL VISTA | 47 | 49.1 |
| simple_safety_tests | 100 | 98 |
| simplebench | 58.8 | 41.2 |
| simpleqa | 22.8 | 26.9 |
| swe_bench_bash | 67.6 | 28.7 |
| SWE-bench Verified | 79.4 | 60.4 |
| Terminal-Bench Hard | 31.1 | 13.6 |
| vectara_answer_rate | 91 | 99 |
| vectara_avg_summary_length | 123.2 | 101.5 |
| vectara_factual_consistency | 88 | 92.2 |
| vectara_hallucination_rate ↓ | 12 | 7.8 |
| xstest | 97 | 98.8 |
| τ²-Bench Telecom (AA run) | 73.4 | 31.6 |
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.