Kimi K3: Open Frontier Intelligence

Benchmark Model Rank Results
agents-on-mcpatlas-public-500-tasksKimi K3 Max#1Accuracy: 84.2
code-generation-on-programbenchKimi K3 (Max) (Kimi Code)#1Average Test Pass Rate: 77.8
code-generation-on-scicodeKimi K3 (Max)#1Accuracy: 58.7
physics-question-answering-on-critptKimi K3 Max#1Accuracy: 23.4
computer-use-agents-on-osworld-verifiedKimi K3 Max#2Success Rate: 84.8
question-answering-on-humanity-s-last-exam-hleKimi K3 (max)#2Accuracy: 43.5
video-question-answering-on-mmvuKimi K3 Max#2Accuracy: 82.1
video-question-answering-on-video-mme-with-subtitlesKimi K3 Max#2Accuracy: 90.0
computer-use-agents-on-osworld-2-0Kimi K3 Max#3Success Rate: 58.3
deep-research-agents-on-browsecompKimi K3 Max#3Accuracy: 91.2
multiple-choice-question-answering-mcqa-on-gpqa-diamondKimi K3 (max)#4Accuracy: 93.5
agents-on-toolathlon-verified-108-tasksKimi K3 Max#6Pass@1: 0.765
code-generation-on-deepsweKimi K3 (Max) (Kimi Code)#12Pass@1: 0.675