List your company

Multi-round Co-reference Resolution on Long-context benchmarks

38.5Score (8k Context)CLP
23.4227.33531.2535.165Mar 18, 2026

Evaluation results

MethodLinks
CLPBackbone=Qwen3 8B2026.0338.533.931.933.930.1
L2ABackbone=Qwen3 8B2026.0336.334.131.234.626.7
DCABackbone=Qwen3 8B2026.0335.735.533.429.217.2
BaseBackbone=Qwen3 8B2026.0335.734.932.529.98.9
NSABackbone=Qwen 2.5 7B2026.033526.116.613.78.7
CLPBackbone=Qwen 2.5 7B2026.0334.630.428.926.321.3
FlexPrefillBackbone=Qwen3 8B2026.033429.831.931.625.5
L2A (Ours)Backbone=Qwen 2.5 7B2026.0332.931.525.824.219.3
S^2-attnBackbone=Qwen 2.5 7B2026.0332.531.826.925.919
SelfExtendBackbone=Qwen 2.5 7B2026.0332.430.730.82111.4
BaseBackbone=Qwen 2.5 7B2026.0332.431.529.625.811.6
DCABackbone=Qwen 2.5 7B2026.033230.929.922.414.8
FlexPrefillBackbone=Qwen 2.5 7B2026.0329.72728.324.517
S^2-attnBackbone=Qwen 2.5 1.5B2026.0326.822.21917.812.8
DCABackbone=Qwen 2.5 1.5B2026.0326.722.817.613.911
L2ABackbone=Qwen 2.5 1.5B2026.0326.420.71615.69.4
BaseBackbone=Qwen 2.5 1.5B2026.0326.422.71815.45.6
CLPBackbone=Qwen 2.5 1.5B2026.0326.322.819.719.114.3
SelfExtendBackbone=Qwen 2.5 1.5B2026.0326.323.118.618.78.9
NSABackbone=Qwen 2.5 1.5B2026.0325.618.78.39.76.1
FlexPrefillBackbone=Qwen 2.5 1.5B2026.032419.416.51312.1