| Benchmark | GPT-5 | Kimi K3 |
|---|---|---|
| AA Agentic Index | 49.7 | 54.3 |
| AA Intelligence | 35.3 | 60 |
| AA-LCR | 76.3 | 82.7 |
| AA-Omniscience | -8.7 | 46 |
| arc_agi_1 | 56.2 | 65.7 |
| ARC-AGI-2 | 7.5 | 12.4 |
| Artificial Analysis Coding Index | 38.9 | 76.2 |
| browsecomp | 54.9 | 91.2 |
| coding_arena_elo | 1419 | 1674 |
| critpt | 5.7 | 23.4 |
| gdpval | 32.5 | 59.4 |
| GPQA Diamond | 87.3 | 93.5 |
| HLE | 28.5 | 56 |
| matharena_visual_math_overall | 78.8 | 91.3 |
| MMMU-Pro | 78.4 | 83.4 |
| OmniScience Accuracy | 40.3 | 47.6 |
| OmniScience Non-Hallucination | 21 | 49.1 |
| scicode | 43 | 58.7 |
| simplebench | 56.7 | 60.7 |
| simpleqa_verified | 50.6 | 42.7 |
| TauBench V3 - Banking | 22.1 | 46 |
| Terminal-Bench 2.1 | 35.2 | 88.3 |
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.