ResearchBenchmarksOutcome Reasoning on Code-PreferenceOutcome Reasoning on Code-PreferenceFollow77M' (F1 Mean)GPT-545.38453.59261.870.008May 17, 2025Evaluation resultsMethodMethodLinksM' (F1 Mean)Y' (F1 Mean)GPT-5Model=GPT-5Model=GPT-52025.057771GPT-o4Model=GPT-o4Model=GPT-o42025.0574.466.8Llama4-MModel=Llama4-MModel=Llama4-M2025.0562.755.9DeepSeekModel=DeepSeekModel=DeepSeek2025.0558.351.6Qwen3Model=Qwen3Model=Qwen32025.0557.150.4Gemini2.5Model=Gemini2.5Model=Gemini2.52025.0554.748.3Llama4-SModel=Llama4-SModel=Llama4-S2025.0546.640