Moonshot · model
Kimi K3 benchmark scores
As of 2026-10-07, Kimi K3 (Moonshot) has measured scores on 57 benchmarks and estimated scores on 69 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 93.5% | measured |
| AA-HLE | 46.9% | measured |
| AA-SciCode | 59.5% | measured |
| Artificial Analysis Intelligence Index | 43.6% | measured |
| CritPt | 23.4% | measured |
| GDPval-AA | 51.8% | measured |
| GPQA Diamond | 93.5% | measured |
| HLE | 56.0% | measured |
| AA-LCR | 88.7% | measured |
| LiveCodeBench v6 | 93.5% | estimated ± 3.1 pp, medium confidence |
| AA-Omniscience Accuracy | 47.6% | measured |
| SciCode | 51.1% | estimated ± 2.9 pp, medium confidence |
| Terminal-Bench 2.1 | 88.3% | measured |
| BrowseComp | 91.2% | measured |
| HLE w/ tools | 59.3% | estimated ± 4.6 pp, high confidence |
| LongBench v2 | 61.5% | estimated ± 5.9 pp, low confidence |
| VITA-Bench | 49.8% | estimated ± 9.9 pp, low confidence |
| τ²-bench results | 94.8% | estimated ± 2.4 pp, medium confidence |
| AA Coding Index | 76.2% | measured |
| AA-MMMU-Pro | 80.5% | measured |
| APEX-Agents | 37.6% | measured |
| APEX-Agents-AA | 41.3% | measured |
| BioMysteryBench (human-difficult) | 44.6% | estimated ± 5.9 pp, low confidence |
| DeepSearchQA | 95.0% | measured |
| SWE-bench Verified | 81.2% | estimated ± 4.4 pp, high confidence |
| AutomationBench | 30.8% | measured |
| BFCL v4 | 77.6% | estimated ± 2.6 pp, medium confidence |
| CyberGym | 79.7% | estimated ± 3.7 pp, high confidence |
| JobBench | 52.9% | measured |
| SWE-bench Pro | 63.3% | estimated ± 4.5 pp, high confidence |
| τ³-bench results | 69.6% | estimated ± 1.8 pp, low confidence |
| WideResearch | 81.9% | estimated ± 1.2 pp, medium confidence |
| AA Agentic Index | 50.6% | measured |
| MMLU-Pro | 85.7% | estimated ± 2.5 pp, low confidence |
| GPQA | 93.5% | measured |
| Gert Labs | 70.9% | estimated ± 3.0 pp, medium confidence |
| AA-AnalystAgent | 38.8% | measured |
| AA EnterpriseOps-Gym | 45.3% | measured |
| AA Harvey LAB | 94.6% | measured |
| ApprenticeBench | 18.0% | measured |
| ARC-AGI-1 | 94.5% | measured |
| ARC-AGI-2 | 60.4% | measured |
| CursorBench 3.1 | 67.4% | estimated ± 3.4 pp, medium confidence |
| CursorBench 3.2 | 60.8% | measured |
| Terminal-Bench 3.0 | 27.6% | estimated ± 6.9 pp, medium confidence |
| FrontierCode 1.1 Main | 43.6% | estimated ± 1.6 pp, medium confidence |
| FrontierSWE v2 | 25.9% | measured |
| OfficeQA Pro | 63.3% | measured |
| OSWorld-Verified | 78.8% | estimated ± 1.9 pp, medium confidence |
| Vals GPQA Diamond | 92.9% | measured |
| Vals LiveCodeBench | 87.2% | measured |
| Vals MMLU-Pro | 88.0% | measured |
| Vals SWE-bench | 93.4% | measured |
| Terminal-Bench 2.1 (Vals AI) | 80.9% | measured |
| VulcanBench v3 | 73.7% | measured |
| AA AutomationBench | 58.3% | measured |
| GDP.pdf | 22.0% | measured |
| AA ITBench | 47.7% | measured |
| AA Tau3 Banking | 46.0% | measured |
| AA Terminal-Bench 2.1 | 85.0% | measured |
| AA Terminal-Bench 4.0 | 12.6% | measured |
| CursorBench 4.0 | 41.3% | estimated ± 1.6 pp, high confidence |
| CWE-bench v1 | 54.9% | estimated ± 6.0 pp, medium confidence |
| DeepSWE | 67.5% | measured |
| HLE w/o tools | 43.5% | measured |
| OSWorld 2.0 | 19.4% | estimated ± 5.5 pp, low confidence |
| PostTrainBench v1.1 | 32.0% | measured |
| ProgramBench | 77.8% | measured |
| Terminal-Bench 4.0 | 16.5% | estimated ± 5.1 pp, low confidence |
| Terminal-Bench-Science 0.1 | 0.1% | estimated ± 4.1 pp, low confidence |
| Toolathlon-Verified | 73.2% | measured |
| Vibe Code Bench | 72.7% | estimated ± 10.3 pp, low confidence |
| CharXiv | 91.3% | measured |
| CharXiv w/o tools | 84.8% | measured |
| ExploitBench | 32.0% | measured |
| ExploitGym | 17.5% | estimated ± 5.0 pp, low confidence |
| C-Eval | 93.3% | estimated ± 0.8 pp, low confidence |
| Claw-Eval | 66.2% | estimated ± 5.6 pp, low confidence |
| DeepPlanning | 52.0% | estimated ± 13.1 pp, low confidence |
| MathVision | 94.3% | measured |
| MCP Atlas | 84.2% | measured |
| MCP-Tasks | 77.1% | estimated ± 0.6 pp, low confidence |
| MMLU-Redux | 97.2% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro | 81.6% | measured |
| NL2Repo | 57.1% | estimated ± 4.7 pp, high confidence |
| QwenClawBench | 69.4% | estimated ± 2.6 pp, medium confidence |
| ScreenSpot Pro | 82.4% | estimated ± 4.5 pp, medium confidence |
| SuperGPQA | 84.7% | estimated ± 7.5 pp, low confidence |
| Terminal-Bench 2.0 | 80.4% | estimated ± 5.9 pp, low confidence |
| Toolathlon | 55.4% | estimated ± 1.7 pp, medium confidence |
| V* | 100.0% | estimated ± 5.1 pp, low confidence |
| VideoMMMU | 88.2% | estimated ± 1.0 pp, high confidence |
| ERQA | 75.8% | estimated ± 2.4 pp, medium confidence |
| HealthBench Hard | 36.0% | estimated ± 5.1 pp, low confidence |
| MedXpertQA (MM) | 77.9% | estimated ± 2.4 pp, high confidence |
| MedXpertQA (Text) | 65.9% | estimated ± 4.5 pp, medium confidence |
| MMLU-Pro (Arcee) | 88.0% | estimated ± 3.4 pp, low confidence |
| React Native Evals | 100.0% | estimated ± 2.4 pp, low confidence |
| ResearchClawBench | 19.5% | estimated ± 1.8 pp, low confidence |
| SWE-Rebench | 58.1% | estimated ± 8.9 pp, low confidence |
| ARC-AGI-3 | 0.9% | estimated ± 2.1 pp, high confidence |
| BioMysteryBench (human-solvable) | 89.4% | estimated ± 0.6 pp, medium confidence |
| DRACO | 82.8% | estimated ± 3.0 pp, medium confidence |
| FrontierCode 1.1 Extended | 55.1% | estimated ± 1.6 pp, low confidence |
| HealthBench (raw) | 55.2% | estimated ± 6.4 pp, low confidence |
| HealthBench (length-adjusted) | 56.8% | estimated ± 2.5 pp, low confidence |
| HealthBench Professional | 59.5% | estimated ± 1.3 pp, medium confidence |
| HealthBench Professional (raw) | 65.2% | estimated ± 5.0 pp, medium confidence |
| HLE-Verified | 35.3% | estimated ± 0.7 pp, low confidence |
| LABBench2 | 79.3% | estimated ± 1.3 pp, low confidence |
| MMMU | 87.7% | estimated ± 0.9 pp, medium confidence |
| LiveCodeBench | 100.0% | estimated ± 8.9 pp, low confidence |
| Agents' Last Exam | 44.5% | estimated ± 6.3 pp, medium confidence |
| OpenHarmony Bench | 57.3% | measured |
| SEC-Bench Pro | 53.9% | estimated ± 4.3 pp, medium confidence |
| BabyVision | 79.9% | estimated ± 4.3 pp, medium confidence |
| MMVU | 80.0% | estimated ± 2.9 pp, medium confidence |
| SimpleVQA | 73.9% | estimated ± 2.5 pp, medium confidence |
| ZeroBench | 23.0% | measured |
| MMMLU | 90.9% | estimated ± 1.1 pp, low confidence |
| MMLU | 94.6% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro w/ Python | 83.4% | measured |
| MathVision w/ Python | 97.8% | measured |
| RealWorldQA | 88.2% | estimated ± 0.8 pp, medium confidence |
| Video-MME (with subtitle) | 89.7% | estimated ± 0.9 pp, medium confidence |
| OmniDocBench 1.5 | 91.7% | estimated ± 5.9 pp, low confidence |