Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

52.09Readability

Tournament-GRPO

34.732439.238743.74548.2513May 26, 2026
Updated 7d ago

Evaluation Results

MethodLinks
2026.05
52.0952.7152.3252.0452.28
2026.05
51.3951.5150.6950.8251.01
2026.05
50.3150.3449.8849.4949.9
2026.05
50.0850.3249.949.8249.99
2026.05
49.749.5348.6948.0248.79
2026.05
47.947.9147.1145.8446.98
2026.05
47.0147.2546.3645.2246.25
2026.05
46.9247.5946.545.0246.33
2026.05
43.8143.5842.440.9842.4
2026.05
43.2242.0739.8139.0540.61
2026.05
4343.3242.7841.8442.6
2026.05
41.7543.4343.0442.6142.79
2026.05
41.4842.0341.2840.3941.17
2026.05
41.4140.2337.8536.8838.6
2026.05
40.0338.4735.835.5436.97
2026.05
38.9337.7834.5134.5535.95
2026.05
35.434.2131.4229.7632.13