List your company

Reasoning on AIME (pass@1 accuracy)

73.33Pass@1 AccuracyBaseline
38.666847.665956.66565.6641Oct 1, 2025

Evaluation results

MethodLinks
BaselineModel=QwQ-32B, Bit-Width=16-162025.1073.33
ThinKV (k=1024)Model=QwQ-32B, Bit-Width=3.4-3.42025.1070.28
PM-KVQModel=QwQ-32B, Bit-Width=3.5-3.52025.1067.86
KIVIModel=QwQ-32B, Bit-Width=2-22025.1060.56
BaselineModel=R1-Qwen-14B, Bit-Width=16-162025.1053.33
ThinKV (k=1024)Model=R1-Qwen-14B, Bit-Width=3.5-3.52025.1050
PM-KVQModel=R1-Qwen-14B, Bit-Width=3.2-3.22025.1043.33
KIVIModel=R1-Qwen-14B, Bit-Width=2-22025.1040