| Benchmark | DeepSeek-V4-Flash | Kimi K3 |
|---|---|---|
| AA Agentic Index | 61.3 | 54.3 |
| AA Intelligence | 51.8 | 60 |
| AA-LCR | 74.3 | 82.7 |
| AA-Omniscience | -14.3 | 46 |
| Agents' Last Exam | 25.2 | 28.3 |
| aime_2026 | 95.8 | 96.7 |
| arc_agi_1 | 84 | 65.7 |
| ARC-AGI-2 | 46 | 12.4 |
| arena_elo | 1439 | 1489 |
| arena_text_factuality | 1434 | 1471 |
| Artificial Analysis Coding Index | 69.1 | 76.2 |
| AutomationBench (Public) | 25.1 | 30.8 |
| browsecomp | 73.2 | 91.2 |
| coding_arena_elo | 1431 | 1674 |
| critpt | 16.6 | 23.4 |
| cybergym | 76.7 | 80 |
| DeepSWE | 54.4 | 67.5 |
| DSBench-FullStack | 68.7 | 73.7 |
| DSBench-Hard | 59.6 | 63 |
| gdpval | 52.9 | 59.4 |
| GDPval-AA v2 | 1189 | 1687 |
| GPQA Diamond | 90.8 | 93.5 |
| HLE | 38.6 | 56 |
| HLE (wo / w tools) | 51.5 | 56 |
| hmmt_feb_2026 | 94.8 | 97 |
| itbenchSre | 31.5 | 47.7 |
| livebench_agentic_coding | 46.8 | 62.2 |
| livebench_coding | 75 | 81.5 |
| livebench_data_analysis | 79.3 | 78.7 |
| livebench_instruction_following | 65.5 | 71.4 |
| livebench_language | 79.2 | 85.5 |
| livebench_math | 86.8 | 84.4 |
| livebench_reasoning | 86.6 | 90.7 |
| MCP Atlas | 69 | 84.2 |
| OmniScience Accuracy | 40.4 | 47.6 |
| OmniScience Non-Hallucination | 15.6 | 49.1 |
| scicode | 49.9 | 58.7 |
| simplebench | 46.3 | 60.7 |
| simpleqa_verified | 45 | 42.7 |
| Tau 3 Banking | 22.9 | 33.4 |
| TauBench V3 - Banking | 39.4 | 46 |
| Terminal-Bench 2.1 | 82.7 | 88.3 |
| toolathlon | 70.3 | 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.