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

Multi-turn Dialogue on MT-Bench (Acceptance Length τ and Speedup Ratio)

5.78Acceptance Length (τ)

PPOW

3.92884.40944.895.3706May 14, 2026
Updated 19d ago

Evaluation Results

MethodLinks
2026.05
5.783.54
2026.05
5.583.02
2026.05
5.532.91
2026.05
5.472.72
2026.05
5.453.73
2026.05
5.253.47
2026.05
5.142.58
2026.05
5.143.52
2026.05
5.123.63
2026.05
5.083.6
2026.05
4.952.64
2026.05
4.952.81
2026.05
4.823.4
2026.05
4.722.82
2026.05
4.73.38
2026.05
4.512.6
2026.05
4.172.35
2026.05
4.122.29
2026.05
4.032.24
2026.05
42.41