| Benchmark | Gemini 3.6 Flash | Kimi K2.5 |
|---|---|---|
| AA Agentic Index | 40.5 | 52.8 |
| AA Intelligence | 52 | 36 |
| AA-LCR | 79 | 73 |
| AA-Omniscience | 22.1 | -7.3 |
| aime_2026 | 96.7 | 95.8 |
| arc_agi_1 | 34.5 | 65.3 |
| ARC-AGI-2 | 2.6 | 11.8 |
| arena_elo | 1481 | 1450 |
| arena_text_factuality | 1470 | 1445 |
| Artificial Analysis Coding Index | 69.2 | 46.8 |
| charxiv_rq | 85.2 | 78.7 |
| coding_arena_elo | 1538 | 1436 |
| critpt | 10.6 | 3.1 |
| gdpval | 46.1 | 38.3 |
| GDPval-AA v2 | 1421 | 1009 |
| GPQA Diamond | 92.8 | 87.9 |
| HLE | 40.8 | 50.2 |
| hmmt_feb_2026 | 89.4 | 87.1 |
| LVBench | 84.2 | 75.9 |
| matharena_visual_math_overall | 87.8 | 80.6 |
| MMMU-Pro | 83.2 | 78.5 |
| OmniScience Accuracy | 50 | 35.2 |
| OmniScience Non-Hallucination | 44.4 | 50 |
| OSWorld-Verified | 83 | 63.3 |
| scicode | 52.7 | 49 |
| simpleqa_verified | 68.7 | 36.9 |
| SWE-bench Pro | 58.7 | 53.8 |
| TauBench V3 - Banking | 29.9 | 14.2 |
| Terminal-Bench 2.1 | 78 | 45.7 |
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.