List your company

Deep Research Evaluation on Deep Research Bench first training epoch (step 600)

52.09ReadabilityTournament-GRPO
34.732439.238743.74548.2513May 26, 2026

Evaluation results

MethodLinks
Tournament-GRPOG=2, Kwin=1, Kfinal=1, M=32026.0552.0952.7152.3252.0452.28
Tournament-GRPOG=2, Kwin=1, Kfinal=1, M=22026.0551.3951.5150.6950.8251.01
Tournament-GRPOG=4, Kwin=2, Kfinal=2, M=42026.0550.3150.3449.8849.4949.9
Tournament-GRPOG=4, Kwin=2, Kfinal=2, M=22026.0550.0850.3249.949.8249.99
Tournament-GRPOG=4, Kwin=2, Kfinal=2, M=12026.0549.749.5348.6948.0248.79
Tournament-GRPOG=2, Kwin=1, Kfinal=1, M=82026.0547.947.9147.1145.8446.98
PairwiseGroup=Baseline2026.0547.0147.2546.3645.2246.25
Tournament-GRPOG=2, Kwin=1, Kfinal=1, M=42026.0546.9247.5946.545.0246.33
Tournament-GRPOG=2, Kwin=1, Kfinal=1, M=12026.0543.8143.5842.440.9842.4
GDPOGroup=Baseline2026.0543.2242.0739.8139.0540.61
Tournament-GRPOG=4, Kwin=2, Kfinal=2, M=32026.054343.3242.7841.8442.6
GRPO explicitGroup=Baseline2026.0541.7543.4343.0442.6142.79
Tournament-GRPOG=4, Kwin=2, Kfinal=2, M=82026.0541.4842.0341.2840.3941.17
GRPO implicitGroup=Baseline2026.0541.4140.2337.8536.8838.6
Dr. GRPO explicitGroup=Baseline2026.0540.0338.4735.835.5436.97
Dr. GRPO implicitGroup=Baseline2026.0538.9337.7834.5134.5535.95
Zero-shotGroup=Baseline2026.0535.434.2131.4229.7632.13