ResearchBenchmarksAbstract Reasoning on ARC (8 held-out tasks)Follow80Success RateSleep-3.218.44061.6Jun 2, 2026Evaluation ResultsMethodMethodLinksSuccess RateSleepBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Learning Setting=Few-shot2026.0680SEALBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Learning Setting=Few-shot2026.0672.5TTTBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Learning Setting=Few-shot2026.0610ICLBackbone=Llama-3.2-1B,...Backbone=Llama-3.2-1B, Learning Setting=Few-shot2026.060