| Benchmark | Gemini 2.5 Flash | GPT-5 |
|---|---|---|
| AA Agentic Index | 18.8 | 49.7 |
| AA Intelligence | 17.7 | 35.3 |
| AA-LCR | 65.7 | 76.3 |
| AA-Omniscience | -29.8 | -8.7 |
| aider_polyglot | 61.9 | 88 |
| AIME | 56.3 | 7.6 |
| AIME 2025 | 72 | 95 |
| AIR-Bench 2024 | 68.7 | 87.7 |
| arc_agi_1 | 33.3 | 56.2 |
| ARC-AGI-2 | 2.5 | 7.5 |
| arena_vision | 1214 | 1212 |
| Artificial Analysis Coding Index | 22.2 | 38.9 |
| critpt | 1.4 | 5.7 |
| frontiermath_tier_4 | 4.2 | 12.5 |
| gdpval | 11.9 | 32.5 |
| GPQA Diamond | 82.8 | 87.3 |
| harmbench | 62.6 | 70 |
| HLE | 12.1 | 28.5 |
| hmmt_feb_2025 | 64.2 | 88.3 |
| IFBench | 50.3 | 73.1 |
| LiveCodeBench | 76.2 | 85 |
| MMMU | 79.7 | 84.2 |
| MMMU-Pro | 69.1 | 78.4 |
| OmniScience Accuracy | 26.1 | 40.3 |
| OmniScience Non-Hallucination | 24.7 | 21 |
| scicode | 35.9 | 43 |
| SEAL VISTA | 49.1 | 49.7 |
| simplebench | 41.2 | 56.7 |
| swe_bench_bash | 28.7 | 65 |
| SWE-bench Verified | 60.4 | 74.9 |
| Terminal-Bench Hard | 13.6 | 37.9 |
| vectara_answer_rate | 99 | 99.9 |
| vectara_avg_summary_length | 101.5 | 162.7 |
| vectara_factual_consistency | 92.2 | 85.3 |
| vectara_hallucination_rate ↓ | 7.8 | 14.7 |
| VisualToolBench | 4.7 | 18.7 |
| τ²-Bench Telecom (AA run) | 31.6 | 86.5 |
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.