ResearchBenchmarksScientific Reasoning on PIQA (test)Follow90.8AccuracyRCSD89.13689.5689090.432Jun 17, 2026Evaluation ResultsMethodMethodLinksAccuracyRCSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0690.8GRPOBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=5002026.0690.3Rubric-GRPOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0690.3Qwen3-8BBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0690.2OPSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0690.1SFTBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0689.2