ResearchBenchmarksMolecular optimization on SYNTHMolecular optimization on SYNTHFollow7,371,212YieldIPS-GRPO-251,295.361,727,624.823,706,5455,685,465.18Jan 29, 2026Evaluation resultsMethodMethodLinksYieldOB100IPS-GRPOThreshold=0.8Threshold=0.82026.017,371,21255,434REINVENTThreshold=0.8Threshold=0.82026.015,695,17360,541IPS-GRPOThreshold=0.85Threshold=0.852026.01158,058117,843REINVENTThreshold=0.85Threshold=0.852026.0141,878293,468