| Benchmark | Gemini 2.5 Flash | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 18.8 | 58.2 |
| AA Intelligence | 17.7 | 53.1 |
| AA-LCR | 65.7 | 77.7 |
| AA-Omniscience | -29.8 | 5.8 |
| arc_agi_1 | 33.3 | 93.7 |
| ARC-AGI-2 | 2.5 | 74 |
| arena_text_factuality | 1437 | 1476 |
| arena_vision | 1214 | 1277 |
| Artificial Analysis Coding Index | 22.2 | 71.1 |
| critpt | 1.4 | 23.4 |
| frontiermath_tier_4 | 4.2 | 27.1 |
| gdpval | 11.9 | 50.1 |
| GPQA Diamond | 82.8 | 93 |
| HLE | 12.1 | 43.7 |
| IFBench | 50.3 | 73.9 |
| MMMU-Pro | 69.1 | 81.2 |
| OmniScience Accuracy | 26.1 | 50.9 |
| OmniScience Non-Hallucination | 24.7 | 17.4 |
| scicode | 35.9 | 56.6 |
| SEAL VISTA | 49.1 | 50.9 |
| Terminal-Bench Hard | 13.6 | 57.6 |
| vectara_answer_rate | 99 | 99.9 |
| vectara_avg_summary_length | 101.5 | 81.7 |
| vectara_factual_consistency | 92.2 | 93 |
| vectara_hallucination_rate ↓ | 7.8 | 7 |
| VisualToolBench | 4.7 | 29.2 |
| τ²-Bench Telecom (AA run) | 31.6 | 87.1 |
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.