ResearchBenchmarksCode Generation on MBPP+ (avg@4)Follow70.8Success Rate (avg@4)SHEAR57.90461.25264.667.948Apr 25, 2026Evaluation ResultsMethodMethodLinksSuccess Rate (avg@4)SHEARBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0470.8GRPOBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0469.3SHEARBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0469GRPOBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0468.8GRPOBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0459.7SHEARBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0458.4