ResearchBenchmarksLong-Context Stability & Retrieval on RULER Tier 2Long-Context Stability & Retrieval on RULER Tier 2Follow76.4SRHalo15.87231.58647.363.014Feb 22, 2026Evaluation resultsMethodMethodLinksSRRTOHaloBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0276.41.29CoT-ValveBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0265.80.85ToTBackbone=LLaMA-3-70B,...Backbone=LLaMA-3-70B, b=5, d=32026.0225.73.5AdaCoTBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0225.11.45CoT-SCBackbone=LLaMA-3-70B,...Backbone=LLaMA-3-70B, k=102026.0221.42.15Standard CoTBackbone=LLaMA-3-70BBackbone=LLaMA-3-70B2026.0218.21