ResearchBenchmarksMathematical & Symbolic Reasoning on Omni-MATH Tier 2Follow42.7Success Rate (SR)Halo11.29219.44627.635.754Feb 22, 2026Evaluation ResultsMethodMethodLinksSuccess Rate (SR)Runtime Overhead (RTO)HaloBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0242.71.29CoT-ValveBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0240.20.85AdaCoTBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0221.51.45ToTBackbone=LLaMA-3-70B,...Backbone=LLaMA-3-70B, b=5, d=32026.0221.33.5CoT-SCBackbone=LLaMA-3-70B,...Backbone=LLaMA-3-70B, k=102026.0215.82.15Standard CoTBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0212.51