| Benchmark | Claude Sonnet 4 | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 39.2 | 49.7 |
| AA Intelligence | 29.8 | 38.7 |
| AA-LCR | 69.7 | 73 |
| AA-Omniscience | 0.2 | 10.1 |
| AIME 2025 | 74 | 99.7 |
| arc_agi_1 | 63.7 | 84.7 |
| ARC-AGI-2 | 13.6 | 33.6 |
| arena_vision | 1207 | 1260 |
| Artificial Analysis Coding Index | 37.6 | 42.6 |
| critpt | 1.1 | 8.6 |
| frontiermath_tier_4 | 0 | 4.2 |
| gdpval | 31.2 | 35.2 |
| GPQA Diamond | 78 | 90.4 |
| harmbench | 98 | 70 |
| HLE | 10.7 | 43.5 |
| IFBench | 55 | 78 |
| livebench | 74.8 | 72.4 |
| mmmlu | 86.5 | 91.8 |
| MMMU-Pro | 62.4 | 81.2 |
| OmniScience Accuracy | 22.7 | 53.4 |
| OmniScience Non-Hallucination | 70.9 | 7.6 |
| OSWorld-Verified | 42.2 | 65.1 |
| scicode | 40 | 50.6 |
| simplebench | 45.5 | 61.1 |
| simpleqa | 15.9 | 68.7 |
| swe_bench_bash | 64.9 | 75.8 |
| SWE-bench Multilingual | 56.9 | 72.7 |
| SWE-bench Pro | 42.7 | 49.6 |
| SWE-bench Verified | 80.2 | 78 |
| TauBench V3 - Banking | 16.7 | 20.8 |
| Terminal-Bench 2.1 | 36.3 | 58 |
| Terminal-Bench Hard | 31.1 | 38.6 |
| vectara_answer_rate | 98.6 | 99.8 |
| vectara_avg_summary_length | 145.8 | 90.2 |
| vectara_factual_consistency | 89.7 | 86.5 |
| vectara_hallucination_rate ↓ | 10.3 | 13.5 |
| τ²-Bench Telecom (AA run) | 64.6 | 80.4 |
| τ³-Bench | 13.8 | 17.5 |
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.