ResearchBenchmarksCode Generation on LiveCodeBench (avg@4)Follow21.1Average@4 ScoreSHEAR17.25218.25119.2520.249Apr 25, 2026Evaluation ResultsMethodMethodLinksAverage@4 ScoreSHEARBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0421.1GRPOBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0419.7SHEARBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0418.3GRPOBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0417.8SHEARBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0417.7GRPOBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0417.4