| Benchmark | Gemini 3.6 Flash | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 40.5 | 58.2 |
| AA Intelligence | 52 | 53.1 |
| AA-LCR | 79 | 77.7 |
| AA-Omniscience | 22.1 | 5.8 |
| aime_2026 | 96.7 | 99.2 |
| arc_agi_1 | 34.5 | 93.7 |
| ARC-AGI-2 | 2.6 | 74 |
| arena_elo | 1481 | 1477 |
| arena_text_factuality | 1470 | 1476 |
| Artificial Analysis Coding Index | 69.2 | 71.1 |
| charxiv_rq | 85.2 | 82.8 |
| coding_arena_elo | 1538 | 1457 |
| critpt | 10.6 | 23.4 |
| DeepSWE 1.1 | 49 | 52 |
| gdpval | 46.1 | 50.1 |
| GPQA Diamond | 92.8 | 93 |
| HLE | 40.8 | 43.7 |
| hmmt_feb_2026 | 89.4 | 97.7 |
| livebench_agentic_coding | 43.4 | 53.8 |
| livebench_coding | 77.9 | 77.5 |
| livebench_data_analysis | 63 | 79.3 |
| livebench_instruction_following | 75.4 | 70.2 |
| livebench_language | 83.9 | 82.6 |
| livebench_math | 86.4 | 94.2 |
| livebench_reasoning | 85.2 | 88.1 |
| matharena_visual_math_overall | 87.8 | 92.5 |
| MMMU-Pro | 83.2 | 81.2 |
| OmniScience Accuracy | 50 | 50.9 |
| OmniScience Non-Hallucination | 44.4 | 17.4 |
| OSWorld-Verified | 83 | 75 |
| scicode | 52.7 | 56.6 |
| simpleqa_verified | 68.7 | 45.3 |
| SWE-bench Pro | 58.7 | 59.1 |
| TauBench V3 - Banking | 29.9 | 39.6 |
| Terminal-Bench 2.1 | 78 | 78.3 |
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.