ResearchBenchmarksMathematical Reasoning on AIME 2024 (Accuracy, Length, Completion Rate)Follow76.7AccuracyASAG38.53248.44158.3568.259Jun 13, 2026Evaluation ResultsMethodMethodLinksAccuracyOutput LengthCompletion RateASAGBackbone=Qwen3-32BBackbone=Qwen3-32B2026.0676.78,46974.8VanillaBackbone=Qwen3-32BBackbone=Qwen3-32B2026.0673.311,325100ASAGBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Llama-8B2026.0646.79,23668.6VanillaBackbone=DeepSeek-R1-D...Backbone=DeepSeek-R1-Distill-Llama-8B2026.064013,472100