| Benchmark | Gemini 3.6 Flash | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 40.5 | 60.2 |
| AA Intelligence | 52 | 43.3 |
| AA-LCR | 79 | 79.3 |
| AA-Omniscience | 22.1 | 0.3 |
| aime_2026 | 96.7 | 98.3 |
| arc_agi_1 | 34.5 | 86.2 |
| ARC-AGI-2 | 2.6 | 52.9 |
| arena_elo | 1481 | 1437 |
| arena_text_factuality | 1470 | 1442 |
| Artificial Analysis Coding Index | 69.2 | 48.7 |
| charxiv_rq | 85.2 | 82.1 |
| coding_arena_elo | 1538 | 1418 |
| critpt | 10.6 | 11.6 |
| gdpval | 46.1 | 48.3 |
| GPQA Diamond | 92.8 | 92.4 |
| HLE | 40.8 | 45.5 |
| hmmt_feb_2026 | 89.4 | 97 |
| livebench_agentic_coding | 43.4 | 50.3 |
| livebench_coding | 77.9 | 76.1 |
| livebench_data_analysis | 63 | 78.2 |
| livebench_instruction_following | 75.4 | 61.8 |
| livebench_language | 83.9 | 79.8 |
| livebench_math | 86.4 | 93.2 |
| livebench_reasoning | 85.2 | 83.2 |
| matharena_visual_math_overall | 87.8 | 86.5 |
| MMMU-Pro | 83.2 | 79.5 |
| mrcr_v2_8needle_128k_average | 91.8 | 83.8 |
| OmniScience Accuracy | 50 | 44.3 |
| OmniScience Non-Hallucination | 44.4 | 38.4 |
| scicode | 52.7 | 52.1 |
| simpleqa_verified | 68.7 | 38.9 |
| SWE-bench Pro | 58.7 | 55.6 |
| TauBench V3 - Banking | 29.9 | 11.1 |
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.