ResearchBenchmarksScientific Reasoning Domain AverageFollow65.2AccuracyDASD62.39263.12163.8564.579May 21, 2026Evaluation ResultsMethodMethodLinksAccuracyDASDBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0565.2OPSDBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0564.3GRPOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0563BaseBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0562.5