ResearchBenchmarksMathematical Reasoning on GSM8K (Strict Accuracy, AL)Follow95.3Strict AccuracySpS94.05294.37694.795.024Apr 5, 2026Evaluation ResultsMethodMethodLinksStrict AccuracyAcceptance Length (AL)SpSm=10, Verifier model=Q...m=10, Verifier model=Qwen 3 32B, Drafter model=Qwen 3 1.7B2026.0495.35.03Cactusm=10, δ=1, Verifier mo...m=10, δ=1, Verifier model=Qwen 3 32B, Drafter model=Qwen 3 1.7B2026.0494.47.13TASm=10, Verifier model=Q...m=10, Verifier model=Qwen 3 32B, Drafter model=Qwen 3 1.7B2026.0494.17.02