| Benchmark | DeepSeek-V3.2 | Kimi K3 |
|---|---|---|
| AA Agentic Index | 39.8 | 54.3 |
| AA Intelligence | 32.8 | 60 |
| AA-LCR | 70.7 | 82.7 |
| AA-Omniscience | -22.5 | 46 |
| aime_2026 | 95.1 | 96.7 |
| apexAgents | 14.5 | 41.3 |
| arc_agi_1 | 57 | 65.7 |
| ARC-AGI-2 | 4 | 12.4 |
| arena_text_factuality | 1433 | 1471 |
| Artificial Analysis Coding Index | 44.2 | 76.2 |
| browsecomp | 51.4 | 91.2 |
| coding_arena_elo | 1368 | 1674 |
| critpt | 2.9 | 23.4 |
| cybergym | 17.3 | 80 |
| deepsearchqa_f1 | 60.9 | 95 |
| gdpval | 18.8 | 59.4 |
| GPQA Diamond | 84 | 93.5 |
| HLE | 40.8 | 56 |
| hmmt_feb_2026 | 84.1 | 97 |
| MCP Atlas | 62.2 | 84.2 |
| OmniScience Accuracy | 33 | 47.6 |
| OmniScience Non-Hallucination | 17.3 | 49.1 |
| ResearchRubrics | 55.8 | 76.2 |
| scicode | 39 | 58.7 |
| simpleqa_verified | 27.5 | 42.7 |
| TauBench V3 - Banking | 18.8 | 46 |
| Terminal-Bench 2.1 | 46.8 | 88.3 |
| toolathlon | 35.2 | 73.2 |
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.