List your company

Mathematical Reasoning on AIME (Accuracy, TPR, Efficiency)

237.5TPR (s)ThinKV
236.616242.583248.55254.517Oct 1, 2025

Evaluation results

MethodLinks
ThinKVModel=R1-Llama-8B, Token Budget=5122025.10237.546.70.21
R-KV (ovl)Model=R1-Llama-8B, Token Budget=5122025.10240.8400.17
R-KV (seq)Model=R1-Llama-8B, Token Budget=5122025.10242.6400.17
ThinKVModel=R1-Llama-8B, Token Budget=10242025.10243.6500.22
R-KV (ovl)Model=R1-Llama-8B, Token Budget=10242025.1024646.70.2
R-KV (seq)Model=R1-Llama-8B, Token Budget=10242025.10247.846.70.2
ThinKVModel=R1-Llama-8B, Token Budget=20482025.10251500.21
R-KV (ovl)Model=R1-Llama-8B, Token Budget=20482025.10253.7500.2
R-KV (seq)Model=R1-Llama-8B, Token Budget=20482025.10254.2500.2
FullKVModel=R1-Llama-8B, Token Budget=–2025.10259.6500.2