ResearchBenchmarksCoding Reasoning on LCBFollow53.4Avg@4RePro49.65650.62851.652.572Dec 1, 2025Evaluation ResultsMethodMethodLinksAvg@4ReProBackbone=Qwen3-8B, Bas...Backbone=Qwen3-8B, Base Training=GRPO2025.1253.4GRPOBackbone=Qwen3-8BBackbone=Qwen3-8B2025.1252.2OriginalBackbone=Qwen3-8BBackbone=Qwen3-8B2025.1249.8