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

Repository-level paper reproduction on PaperBench Code lite (dev)

49.26MU-DPO

PaperCoder

11.50821.30931.1140.911Oct 20, 2025
Updated 1mo ago

Evaluation Results

MethodLinks
2025.10
49.2659.1973.0360.6859.5360.34
2025.10
47.431.7826.5738.4431.8935.22
2025.10
46.4853.9952.0863.1350.3653.21
2025.10
43.736.2823.9129.1526.531.91
2025.10
43.2443.2651.1861.1262.3752.23
2025.10
39.1439.1424.4933.9721.3831.62
2025.10
37.2227.2620.8222.8614.6724.57
2025.10
33.0540.5517.2231.5617.0827.89
2025.10
23.1545.752.4850.3739.8442.31
2025.10
22.2221.3828.7731.2819.3524.6
2025.10
16.231.1931.0935.321.3227.02
2025.10
12.9622.6318.2420.8214.8217.89