ResearchBenchmarksMathematical Reasoning on GSM8K (Acc, FLOPs, A/C)Follow94.6Accuracy (GSM8K)7B Only84.82487.36289.992.438May 7, 2026Evaluation ResultsMethodMethodLinksAccuracy (GSM8K)FLOPs (GSM8K)A/C Ratio (GSM8K)7B OnlyExternal Models=N/AExternal Models=N/A2026.0594.64.4221.4OursExternal Models=falseExternal Models=false2026.0594.52.0346.55RSDExternal Models=trueExternal Models=true2026.0594.42.2741.59STEERExternal Models=falseExternal Models=false2026.0594.42.6635.491.5B OnlyExternal Models=N/AExternal Models=N/A2026.0585.20.97387.56SpecReasonExternal Models=falseExternal Models=false2026.0585.21.0382.72