benchgap
Moonshot · model

Kimi K3 (max) benchmark scores

As of 2026-10-07, Kimi K3 (max) (Moonshot) has measured scores on 15 benchmarks and estimated scores on 1 more.

BenchmarkScoreSource
Terminal-Bench 2.185.0%measured
Terminal-Bench 4.012.6%measured
Terminal-Bench Science 0.12.6%estimated ± 6.5 pp, medium confidence
GPQA Diamond93.5%measured
SciCode59.5%measured
CritPt23.4%measured
GDP.pdf22.0%measured
τ³-Bench Banking46.0%measured
MMMU-Pro81.0%measured
AA-AnalystAgent38.8%measured
Harvey LAB94.6%measured
Humanity's Last Exam46.9%measured
AutomationBench58.3%measured
EnterpriseOps-Gym45.3%measured
APEX-Agents41.3%measured
ITBench SRE47.7%measured