| Benchmark | Gemini 2.5 Flash | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 18.8 | 60.2 |
| AA Intelligence | 17.7 | 43.3 |
| AA-LCR | 65.7 | 79.3 |
| AA-Omniscience | -29.8 | 0.3 |
| AIME 2025 | 72 | 100 |
| arc_agi_1 | 33.3 | 86.2 |
| ARC-AGI-2 | 2.5 | 52.9 |
| arena_text_factuality | 1437 | 1442 |
| arena_vision | 1214 | 1245 |
| Artificial Analysis Coding Index | 22.2 | 48.7 |
| critpt | 1.4 | 11.6 |
| frontiermath_tier_4 | 4.2 | 18.8 |
| gdpval | 11.9 | 48.3 |
| GPQA Diamond | 82.8 | 92.4 |
| HLE | 12.1 | 45.5 |
| hmmt_feb_2025 | 64.2 | 100 |
| IFBench | 50.3 | 75.4 |
| LiveCodeBench | 76.2 | 89 |
| MMMU-Pro | 69.1 | 79.5 |
| OmniScience Accuracy | 26.1 | 44.3 |
| OmniScience Non-Hallucination | 24.7 | 38.4 |
| PropensityBench | 68 | 34.4 |
| scicode | 35.9 | 52.1 |
| SEAL VISTA | 49.1 | 46.6 |
| simplebench | 41.2 | 45.8 |
| swe_bench_bash | 28.7 | 72.8 |
| SWE-bench Verified | 60.4 | 80 |
| Terminal-Bench Hard | 13.6 | 62.2 |
| vectara_answer_rate | 99 | 100 |
| vectara_avg_summary_length | 101.5 | 186.3 |
| vectara_factual_consistency | 92.2 | 89.2 |
| vectara_hallucination_rate ↓ | 7.8 | 10.8 |
| τ²-Bench Telecom (AA run) | 31.6 | 98.7 |
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.