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

Reward Model Evaluation on RewardBench (Spearman ρ)

0.542Spearman ρ

prob-weighted EV (M2)

Updated 16d ago

Evaluation Results

Method	Links
prob-weighted EV (M2) 2026.05		0.542
zero-shot BDAS-1D (M4) 2026.05		0.537
prompted argmax (M1) 2026.05		0.522
supervised residual probe (M3) 2026.05		0.487
prob-weighted EV (M2) 2026.05		0.432
zero-shot BDAS-1D (M4) 2026.05		0.425
prompted argmax (M1) 2026.05		0.422
prob-weighted EV (M2) 2026.05		0.418
prompted argmax (M1) 2026.05		0.413
supervised residual probe (M3) 2026.05		0.398
supervised residual probe (M3) 2026.05		0.375
prob-weighted EV (M2) 2026.05		0.367
zero-shot BDAS-1D (M4) 2026.05		0.345
zero-shot BDAS-1D (M4) 2026.05		0.338
prompted argmax (M1) 2026.05		0.327
supervised residual probe (M3) 2026.05		0.32
zero-shot BDAS-1D (M4) 2026.05		0.298
supervised residual probe (M3) 2026.05		0.291
prompted argmax (M1) 2026.05		0.276
prob-weighted EV (M2) 2026.05		0.241