List your company

Scientific Reasoning on GPQA-D (pass@1 Accuracy, Standard Error)

59.1Pass@1 AccuracyR-KV
48.07650.93853.856.662Jun 2, 2026

Evaluation results

MethodLinks
R-KVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0659.13
VASE-ATTNVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0657.43
VASE-DKVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0656.83
CurDKVBase Model=Qwen3-14B, Cache Compression=~ 4x2026.0648.53.1