| Benchmark | DeepSeek-V4-Pro | Kimi K3 |
|---|---|---|
| AA Agentic Index | 63.3 | 54.3 |
| AA Intelligence | 53 | 60 |
| AA-LCR | 70 | 82.7 |
| AA-Omniscience | -10.6 | 46 |
| Agents' Last Exam | 16.5 | 28.3 |
| aime_2026 | 96.7 | 96.7 |
| apexAgents | 24.3 | 41.3 |
| arena_elo | 1457 | 1489 |
| arena_text_factuality | 1450 | 1471 |
| Artificial Analysis Coding Index | 59.4 | 76.2 |
| AutomationBench (Public) | 12.8 | 30.8 |
| browsecomp | 83.4 | 91.2 |
| coding_arena_elo | 1582 | 1674 |
| critpt | 13 | 23.4 |
| cybergym | 52.7 | 80 |
| DeepSWE | 12.8 | 67.5 |
| DSBench-FullStack | 41.8 | 73.7 |
| DSBench-Hard | 31.1 | 63 |
| gdpval | 49 | 59.4 |
| GDPval-AA v2 | 1307 | 1687 |
| GPQA Diamond | 90.5 | 93.5 |
| HLE | 48.2 | 56 |
| HLE (wo / w tools) | 48.2 | 56 |
| hmmt_feb_2026 | 95.2 | 97 |
| itbenchSre | 38.3 | 47.7 |
| livebench_agentic_coding | 42.6 | 62.2 |
| livebench_coding | 70 | 81.5 |
| livebench_data_analysis | 74.5 | 78.7 |
| livebench_instruction_following | 62.4 | 71.4 |
| livebench_language | 78.1 | 85.5 |
| livebench_math | 90.7 | 84.4 |
| livebench_reasoning | 82.7 | 90.7 |
| MCP Atlas | 74.2 | 84.2 |
| OmniScience Accuracy | 43 | 47.6 |
| OmniScience Non-Hallucination | 12.2 | 49.1 |
| Program Bench | 47.8 | 77.8 |
| scicode | 50 | 58.7 |
| simplebench | 50.9 | 60.7 |
| simpleqa_verified | 57.9 | 42.7 |
| Tau 3 Banking | 26 | 33.4 |
| TauBench V3 - Banking | 30.1 | 46 |
| Terminal-Bench 2.1 | 72.1 | 88.3 |
| toolathlon | 51.8 | 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.