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

Medical Question Answering on HealthBench Medicine N=5,000 (overall)

26.1Rubric Score

RVPO (k=0.5→2.0 / 1.0→2.0)

Updated 26d ago

Evaluation Results

Method	Links
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		26.1
GRPO (Explicit) 2026.05		23.6
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		23.6
GRPO (Implicit) 2026.05		23.4
GRPO (Implicit) 2026.05		23.4
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		23
RVPO-explicit (β=1.0/0.5) 2026.05		22.7
RVPO-min (k=∞) 2026.05		22.5
GRPO (Explicit) 2026.05		22.1
GDPO 2026.05		21.5
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		20.4
GDPO 2026.05		19.8
GRPO (Implicit) 2026.05		19.3
GRPO (Implicit) 2026.05		19.3
GDPO 2026.05		19.2
RVPO-min (k=∞) 2026.05		19.1
GRPO (Explicit) 2026.05		19
RVPO-min (k=∞) 2026.05		19
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		18.9
RVPO-explicit (β=1.0/0.5) 2026.05		18.8
RVPO-explicit (β=1.0/0.5) 2026.05		18.4
RVPO-min (k=∞) 2026.05		18.1
RVPO-explicit (β=1.0/0.5) 2026.05		17.8
RVPO-explicit (β=1.0/0.5) 2026.05		16.3
GRPO (Implicit) 2026.05		15.4
RVPO (k=0.5→2.0 / 1.0→2.0) 2026.05		14.7
RVPO-min (k=∞) 2026.05		12.4
GDPO 2026.05		11.7
GRPO (Explicit) 2026.05		10.2
GRPO (Implicit) 2026.05		7.2
GRPO (Explicit) 2026.05		5.3
GDPO 2026.05		2.6
RVPO-explicit (β=1.0/0.5) 2026.05		1.1
RVPO-min (k=∞) 2026.05		0
GRPO (Explicit) 2026.05		0
GDPO 2026.05		0