Apple: Recursive Language Models for Long-Context Processing with Self-Reflection

Apple ML Research · rss · 2026-07-09

Apple ML Research published a paper examining the performance of Recursive Language Models (RLMs) in processing long contexts.

The article points out that even with expanded context windows, models still struggle to extract and reason over information in long texts. RLMs address this by breaking down long contexts into recursive sub-queries agentic-style during inference. This study deeply explores how to effectively select trajectories for these context-interaction programs, validating the surprising effectiveness of self-reflective program search in tackling long-context challenges.

Original post →

More from Research

Research channel →