ResearchBenchmarksLong-context Question Answering on InfiniteBench long-context EN.QALong-context Question Answering on InfiniteBench long-context EN.QAFollow53.9AccuracyCGLS22.28430.49238.746.908Jun 13, 2025Evaluation resultsMethodMethodLinksAccuracyCGLSBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Token Budget=20B2025.0653.9BL-CBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Token Budget=20B2025.0640.1BL-RBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Token Budget=20B2025.0623.5