ResearchBenchmarksCode Generation on HumanEval+ (avg@4)Follow79.6Pass Rate @4SHEAR59.21664.50869.875.092Apr 25, 2026Evaluation ResultsMethodMethodLinksPass Rate @4SHEARBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0479.6GRPOBackbone=Qwen2.5-14B-BaseBackbone=Qwen2.5-14B-Base2026.0479.3SHEARBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0472.1GRPOBackbone=Qwen2.5-Coder-7BBackbone=Qwen2.5-Coder-7B2026.0471SHEARBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0463.5GRPOBackbone=Llama3.1-8B-I...Backbone=Llama3.1-8B-Instruct2026.0460