Rankings / Moonshot AI
Kimi K2 5
released 2026-01-27
BenchAtlas Index
as of 2026-07-14
thinking
rank #160
8 families · 4 categories · medium
thinking
rank #193
9 families · 4 categories · medium
base configuration
rank #219
7 families · 5 categories · medium
Benchmark evidence
39 results
agentic coding
| τ²-Bench | thinking | 95.9 % | independent | Artificial Analysis |
| τ²-Bench | no reasoning | 81.3 % | independent | Artificial Analysis |
| τ²-Bench subset=banking | thinking | 14.2 % | independent | Artificial Analysis |
coding
| SciCode | thinking | 49.0 % | independent | Artificial Analysis |
| SciCode | no reasoning | 39.6 % | independent | Artificial Analysis |
external indices
| AA Coding Index | thinking | 46.8 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | thinking | 38.1 points | independent | Artificial Analysis |
| Intelligence Index v4.1 | no reasoning | 29.4 points | independent | Artificial Analysis |
| ECI | 148.2 points | independent | Epoch AI Benchmarking Hub | |
| ECI | no reasoning | 148.2 points | independent | Epoch AI Benchmarking Hub |
factuality
| MASK contamination=Potential contamination warning: This model was evaluated after the public release of MASK, allowing model builder access to the prompts and solutions. | 70.5 % 66.0–75.0 | independent | Scale Labs |
knowledge science
| GPQA Diamond implementation=artificial-analysis | thinking | 87.9 % | independent | Artificial Analysis |
| GPQA Diamond implementation=vals-ai | thinking | 84.1 % | independent | Vals AI |
| GPQA Diamond implementation=artificial-analysis | no reasoning | 78.9 % | independent | Artificial Analysis |
| Humanity's Last Exam implementation=artificial-analysis | thinking | 29.4 % | independent | Artificial Analysis |
| Humanity's Last Exam contamination=Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. · implementation=scale | 24.4 % 22.6–26.2 | independent | Scale Labs | |
| Humanity's Last Exam implementation=artificial-analysis | no reasoning | 12.3 % | independent | Artificial Analysis |
| MMLU-Pro implementation=vals-ai | thinking | 85.9 % | independent | Vals AI |
long context instruction
| AA-LCR | thinking | 65.3 % | independent | Artificial Analysis |
| AA-LCR | no reasoning | 59.0 % | independent | Artificial Analysis |
| IFBench | thinking | 70.2 % | independent | Artificial Analysis |
| IFBench | no reasoning | 43.7 % | independent | Artificial Analysis |
| MultiChallenge | 61.4 % 60.9–61.9 | independent | Scale Labs |
multimodal
| MMMU implementation=vals-ai | thinking | 84.3 % | independent | Vals AI |
| VISTA | 41.9 % 37.6–46.1 | independent | Scale Labs |
professional
| CaseLaw | thinking | 58.7 % | independent | Vals AI |
| CorpFin | thinking | 68.3 % | independent | Vals AI |
| MedQA | thinking | 94.4 % | independent | Vals AI |
| TaxEval | thinking | 74.2 % | independent | Vals AI |
reasoning math
| AIME implementation=vals-ai | thinking | 95.6 % | independent | Vals AI |
| ARC-AGI-1older version split=public_eval · model_type=CoT | 73.1 % | independent | ARC Prize Leaderboard | |
| ARC-AGI-1older version split=semi_private · model_type=CoT | 65.3 % | independent | ARC Prize Leaderboard | |
| ARC-AGI-2 split=public_eval · model_type=CoT | 12.1 % | independent | ARC Prize Leaderboard | |
| ARC-AGI-2 split=semi_private · model_type=CoT | 11.8 % | independent | ARC Prize Leaderboard | |
| ARC-AGI-2 split=public_eval · model_type=CoT | 0.3 usd_per_task | independent | ARC Prize Leaderboard | |
| ARC-AGI-2 split=semi_private · model_type=CoT | 0.3 usd_per_task | independent | ARC Prize Leaderboard | |
| ARC-AGI-1older version split=semi_private · model_type=CoT | 0.1 usd_per_task | independent | ARC Prize Leaderboard | |
| ARC-AGI-1older version split=public_eval · model_type=CoT | 0.1 usd_per_task | independent | ARC Prize Leaderboard | |
| EnigmaEval contamination=Potential contamination warning: This model was evaluated after the public release of EnigmaEval, allowing model builder access to the prompts and solutions. | 3.4 % 2.4–4.4 | independent | Scale Labs |
Agent + model results
systems, not bare-model scores
| agent + model Epoch Inspect harness + Kimi K2 5 | SWE-bench Verified | 73.8 % | independent | Epoch AI Benchmarking Hub |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench Verified | 70.8 % | unverified | SWE-bench Leaderboard |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench bash-only | 70.8 % | unverified | SWE-bench Leaderboard |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench Multilingual | 67.3 % | community | SWE-bench Leaderboard |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench Multilingual | 0.7 usd_per_task | community | SWE-bench Leaderboard |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench bash-only | 0.1 usd_per_task | unverified | SWE-bench Leaderboard |
| agent + model mini-SWE-agent + Kimi K2 5 | SWE-bench Verified | 0.1 usd_per_task | unverified | SWE-bench Leaderboard |
| agent + model Artificial Analysis harness + Kimi K2 5 | Terminal-Bench 2.1 | 45.7 % | independent | Artificial Analysis |
| agent + model terminus-2 + Kimi K2 5 | Terminal-Bench 2.0 | 43.2 % | community | Terminal-Bench Leaderboard |
| agent + model Artificial Analysis harness + Kimi K2 5 | Terminal-Bench Hard | 34.9 % | independent | Artificial Analysis |
| agent + model Artificial Analysis harness + Kimi K2 5 | Terminal-Bench Hard | 18.9 % | independent | Artificial Analysis |
These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.