ResearchBenchmarksScientific Reasoning on RUBRICHUB (test)Follow55.7AccuracyRCSD50.60451.92753.2554.573Jun 17, 2026Evaluation ResultsMethodMethodLinksAccuracyRCSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0655.7OPSDBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=1002026.0654.5Rubric-GRPOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0652.9GRPOBackbone=Qwen3-8B, Tra...Backbone=Qwen3-8B, Training Steps=5002026.0651.9SFTBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0651.2Qwen3-8BBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0650.8