- Terminal-Bench 4.0 (Claude Code) 65.7
- Terminal-Bench-Science 0.1 (Claude Code) 38.1
- SWE Marathon v1.1 (Claude Code) 51.7
- BenchCAD w/ tools (w/o tools 72.8) 92.7
- Program Bench (binary score, >=95% counts as solved) 31.5
- DeepSWE v1.1 (Claude Code) 75.5
- PostTrain Bench v1.2 (H20 GPUs) 36.5
- Kimi Code Bench 2.0 (Internal) 77.1
- Kimi Code Bench 3.0 Long-horizon (Internal) 61.2
- Best of HLE-Full w/ tools 70.0 / HLE-Full 51.1 70
- BrowseComp 92.1
- BioMysteryBench (v11) 80.6
- CharXiv (RQ) 89
- MMMU-Pro 84.1
- GDP.pdf 34.6
- ChartoGraphy w/ python 90.2
- BabyVision w/ python 95.4
- ZeroBench_main (avg@5 / pass@5) 29.8 / 39.0 39
- Agents Last Exam (ALL) (CLI 44.2) 43.2
- Automation Bench-AA 73.3
- GDPval-AA v2.1 (Elo-score) 1846
- AA-Briefcase v1.1 (Elo-score) 1678
- APEX-Agents 67.2
- OfficeQA Pro v2 64.4
- OfficeQA Pro 67.4
- Frontier Finance 58.6
- Harvey Legal Agent (public 1,251 tasks, Stirrup harness, all-pass rate) 23.7
- HANDBOOK.md 44.6
- OSWorld V2.1 79.6
- Business Operation (Internal) 37.8
- DECK-Arena (Internal, Elo-score) 1456
- Design Bench (Internal, Elo-score) 957
- Kimi Hallu Bench Hard w/ tools (Internal, lower is better) 8.3
- GPQA-Diamond 94.2
- PerceptionBench 68.3
- ExploitBench (pass@5 max capability score, 300 turns) 74.7
- ExploitGym (6-hour budget) 20.7
- xdev-bench-hard (Internal, pass@3) 81
Kimi K3.1 Pro
Kimi K3.1 Pro ranks first on 9 of the 58 benchmarks it has scores on.
Moonshot AI Released 23 Oct 2026
70
Benchmarks with a score
9
Ranked first, out of five or more models
17
In the top three
Moonshot AI Oct 2026
Kimi K3.1 Pro
Closed weights
No Epoch Capabilities Index score yet
- MMMU-Pro 84.1% #1 /72
- Agents' Last Exam 43.4% #3 /12
- CritPt 24.6% #21 /160
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Last Exam | 70% | #1 of 147 | Model card | |
| Creative Writing v3 | 83.2 | #1 of 120 | Official leaderboard | |
| Extended NYT Connections | 93.4% | #13 of 97 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 79.9 | #16 of 119 | Official leaderboard | |
| HLE-Diamond | 38.7% | #11 of 18 | Official leaderboard | |
| Design Bench (Moonshot) | 957 | #1 of 1 | Model card | |
| HANDBOOK.md | 44.6% | #1 of 1 | Model card | |
| Kimi Hallu Bench | 8.30% | #1 of 1 | Model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 94.2% | #6 of 284 | Model card | |
| ProofBench | 91% | #9 of 83 | Epoch AI | |
| CritPt | 24.6% | #21 of 160 | Epoch AI | |
| ProofBench v1.1 | 91% | #10 of 51 | Official leaderboard | |
| MysteryMechanism | 26.1% | #16 of 29 | Official leaderboard | |
| BioMysteryBench | 80.6% | #5 of 8 | Model card | |
| ArXivLean | 17% | #5 of 7 | Official leaderboard | |
| Terminal-Bench-Science | 38.1% | #8 of 9 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OpenDesign Arena | 91.0 | #1 of 19 | Official leaderboard | |
| Code Arena | 1,695 | #8 of 124 | Official leaderboard | |
| DeepSWE | 75.5% | #5 of 51 | Model card | |
| ALE-Bench | 1,933 | #13 of 121 | Epoch AI | |
| Vibe Code Bench | 84.9% | #18 of 115 | Official leaderboard | |
| IOI v2 (Vals) | 81.9% | #11 of 45 | Official leaderboard | |
| Kimi Code Bench v2 | 77.1% | #1 of 3 | Model card | |
| SWE-Marathon | 51.7% | #4 of 11 | Model card | |
| ProgramBench | 31.5% | #15 of 20 | Model card | |
| Brokk Power Ranking | 3.30 | #16 of 19 | Official leaderboard | |
| Kimi Code Bench 3.0 Long-horizon | 61.2% | #1 of 1 | Model card | |
| PostTrainBench 1.2 | 36.5% | #1 of 1 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AutomationBench | 73.3% | #1 of 41 | Model card | |
| BrowseComp | 92.1% | #3 of 77 | Model card | |
| Terminal-Bench 4.0 | 65.7% | #3 of 23 | Model card | |
| Agent Arena | 0.06 | #13 of 53 | Official leaderboard | |
| OSWorld 2.1 | 79.6% | #3 of 5 | Model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| APEX-Agents | 67.2% | #1 of 12 | Model card | |
| PharmaBench (micro1) | 27.5% | #1 of 12 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 23.7% | #3 of 17 | Model card | |
| BenchCAD | 92.7% | #2 of 10 | Model card | |
| Agents' Last Exam | 43.4% | #3 of 12 | Official leaderboard | |
| GDPval-AA v2.1 | 1,846 | #2 of 8 | Model card | |
| Vals Index v2.2 | 62.5% | #9 of 32 | Official leaderboard | |
| OfficeQA Pro | 67.4% | #5 of 14 | Model card | |
| Excel Modeling Benchmark v1.1 | 70% | #15 of 37 | Official leaderboard | |
| Legal Research Bench | 37.5% | #42 of 88 | Official leaderboard | |
| Vals Finance Agent v2.1 | 61.5% | #17 of 35 | Official leaderboard | |
| AA-Briefcase v1.1 | 1,678 | #3 of 6 | Model card | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 11.7% | #17 of 33 | Official leaderboard | |
| Tax Agent Bench | 24.9% | #41 of 79 | Official leaderboard | |
| RWE-bench | 1.00% | #6 of 11 | Official leaderboard | |
| Vals Finance Index | 56% | #20 of 35 | Official leaderboard | |
| Vals Legal Index | 67% | #25 of 33 | Official leaderboard | |
| FrontierFinance | 58.6% | #2 of 2 | Model card | |
| Business Operation (Moonshot) | 37.8% | #1 of 1 | Model card | |
| DECK-Arena | 1,456 | #1 of 1 | Model card | |
| OfficeQA Pro V2 | 64.4% | #1 of 1 | Model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU-Pro | 84.1% | #1 of 72 | Model card | |
| BabyVision | 95.4% | #1 of 20 | Model card | |
| Chartography | 90.2% | #1 of 12 | Model card | |
| CharXiv | 89% | #10 of 67 | Model card | |
| VoxelBench | 2,224 | #9 of 56 | Official leaderboard | |
| ZeroBench | 39% | #5 of 25 | Model card | |
| GDP.pdf | 34.6% | #5 of 15 | Model card | |
| PerceptionBench (Moonshot) | 68.3% | #1 of 3 | Model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DeepsecBench | 37.5 | #5 of 28 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 61.7% | #25 of 71 | Official leaderboard | |
| ExploitBench | 74.7% | #8 of 19 | Model card | |
| ExploitGym | 20.7% | #9 of 14 | Model card | |
| xdev-bench-hard | 81% | #1 of 1 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CAIS Text Capabilities Index | 46.4 | #12 of 66 | Official leaderboard | |
| LiveBench | 71.3% | #10 of 28 | Official leaderboard | |
| Vals Science Index | 44.4% | #9 of 23 | Official leaderboard |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.