Kimi K2-Instruct-0905 model card (Hugging Face) Sep 5, 2025
- SWE-Bench verified, ACC 69.2
- SWE-Bench Multilingual, ACC 55.9
- Multi-SWE-Bench, ACC 33.5
- Terminal-Bench, ACC 44.5
- SWE-Dev, ACC 66.6
Kimi K2 Sep 2025
Kimi K2 ranks first on 2 of the 53 benchmarks with five or more models scored.
55
Benchmarks with a score
2
Ranked first, out of five or more models
2
In the top three
Moonshot AI Sep 2025
Kimi K2
Sep 2025 snapshot Open weights Flagship
No Epoch Capabilities Index score yet
- Terminal-Bench 2.0 27.8% #33 /43
- Toolathlon 13% #48 /50
- VITA-Bench 11.5% #17 /18
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 | |
|---|---|---|---|---|
| Slop Score | 18.3 | #1 of 17 | Official leaderboard | |
| Encyclo-K | 44.7% | #4 of 23 | Official leaderboard | |
| LMArena (Elo) | 1,418 | #106 of 335 | Official leaderboard | |
| Fiction.LiveBench | 66.7% | #26 of 57 | Fiction.live leaderboard, via Epoch AI | |
| Longform Creative Writing (EQ-Bench) | 63.8 | #56 of 119 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 2711.0 | #42 of 77 | Official leaderboard | |
| BullshitBench | 20% | #70 of 107 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 6.20% | #106 of 148 | Official leaderboard | |
| HUMAINE | 20 | #44 of 50 | Official leaderboard | |
| BullshitBench V2 | 10% | #114 of 129 | Official leaderboard | |
| AA-LCR Extended | 32.9% | #45 of 49 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 17.9% | #96 of 104 | Official leaderboard | |
| CL-bench | 11.9% | #23 of 24 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ChemBench | 59.7% | #12 of 24 | Official leaderboard | |
| SciArena | 1,002 | #21 of 34 | Official leaderboard | |
| ForecastBench | 0.14 | #53 of 82 | Official leaderboard | |
| AMO-Bench | 7.50% | #18 of 21 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 71.2% | #13 of 60 | Official leaderboard | |
| SWE-fficiency | 5.37% | #11 of 17 | Official leaderboard | |
| AetherCode | 9.80% | #7 of 9 | Official leaderboard | |
| DesignArena Website | 1,082 | #119 of 151 | Official leaderboard | |
| Multi-SWE-bench | 33.5% | #8 of 10 | Lab model card | |
| AInsteinBench | 29.1% | #5 of 6 | Official leaderboard | |
| DesignArena UI Component | 1,021 | #118 of 139 | Official leaderboard | |
| DesignArena Code | 1,024 | #122 of 143 | Official leaderboard | |
| SWE-bench Multilingual | 55.9% | #64 of 73 | Lab model card | |
| DesignArena Game Dev | 964 | #130 of 143 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 0% | #35 of 38 | Official leaderboard | |
| APEX-SWE | 17.4% | #63 of 65 | Official leaderboard | |
| CC-Bench | 14.9% | #4 of 4 | Official leaderboard | |
| SWE-Dev | 66.6% | #1 of 1 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 30 | #39 of 188 | Official leaderboard | |
| Terminal-Bench | 44.5% | #6 of 22 | Lab model card | |
| LLM-WikiRace | 7.00% | #8 of 21 | Official leaderboard | |
| MCP-Universe | 26.4% | #12 of 30 | Official leaderboard | |
| Meta Agents Research Environments (ARE) / Gaia2 | 20.1% | #6 of 13 | Official leaderboard | |
| MCPMark | 21.9% | #14 of 28 | Official leaderboard | |
| CompileBench | 80% | #10 of 20 | Official leaderboard | |
| Terminal-Bench 2.0 | 27.8% | #33 of 43 | Official leaderboard | |
| VITA-Bench | 11.5% | #17 of 18 | Official leaderboard | |
| SnorkelWordle | 54% | #18 of 19 | Official leaderboard | |
| Toolathlon | 13% | #48 of 50 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 11.1% | #43 of 44 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| StockBench | 1.00 | #1 of 13 | Official leaderboard | |
| ProfBench | 44.2% | #50 of 72 | Official leaderboard | |
| PLawBench | 57.3% | #8 of 11 | Official leaderboard | |
| APEX-1 | 52.4% | #65 of 71 | Official leaderboard | |
| GDPval Extended | 0% | #40 of 43 | Official leaderboard | |
| MBABench | 14.4 | #6 of 6 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MineBench | 963 | #69 of 75 | Official leaderboard | |
| CaP-X (CaP-Bench) | 5.17% | #10 of 10 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Spiral-Bench v1.2 | 57.2 | #7 of 21 | Official leaderboard | |
| Spiral-Bench v1.0 | 73.0 | #6 of 13 | Official leaderboard | |
| SusVibes | 9.10% | #9 of 10 | Official leaderboard | |
| Lakera B3 (Backbone Breaker Benchmark) | 41.7% | #32 of 35 | Official leaderboard |
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| SWE-bench Verified | 69.2% SWE-Bench verified, ACC | 71.2% Leaderboard, Lingxi v1.5 | −2.0 pts |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.