1T MoE with 100-agent swarm coordination, 99% HumanEval
SI v3 uses weighted percentages from one Artificial Analysis configuration. Ranking requires all five benchmarks. Partial scores average only available weights. Scores are not comparable to SI v2 or a probability of reaching singularity.
Compared with 51 models released within six months of it, across 6 benchmarks where scores spread out.
| Benchmark | Score | Rank |
|---|---|---|
HumanEval Coding ability - generating correct Python functions | 99% | #1 / 50 |
MATH Competition-level mathematics problems | 98% | #6 / 50 |
MMLU Tests knowledge across 57 subjects from STEM to humanities | 92% | #11 / 54 |
SWE-bench Real-world GitHub issue resolution | 76.8% | #11 / 39 |
Terminal Agentic terminal coding tasks requiring multi-step execution | 50.8% | #17 / 58 |
Arena Elo Human preference ranking via blind comparisons | 1450 | #22 / 51 |
ARC-AGI Novel reasoning tasks requiring fluid intelligence | 12% | #35 / 40 |
GPQA PhD-level science questions even experts struggle with | 87.6% | #56 / 96 |
HLEArtificial Analysis Challenging multidisciplinary questions evaluated by Artificial Analysis | 30.7% | #64 / 100 |
| Benchmark / source | Score |
|---|---|
HLE Artificial AnalysisKimi K2.5 (Reasoning) | 30.7% |
MMMU-Pro Artificial AnalysisKimi K2.5 (Reasoning) | 75.4% |
Terminal-Bench 2.1 Artificial AnalysisKimi K2.5 (Reasoning) | 45.7% |
AA-LCR Artificial AnalysisKimi K2.5 (Reasoning) | 78% |
CritPt Artificial AnalysisKimi K2.5 (Reasoning) | 3.1% |
IFBench Artificial AnalysisKimi K2.5 (Reasoning) | 70.2% |