Recursive Language Models for Long Context
Research Paper#Large Language Models (LLMs), Long Context, Recursive Processing🔬 Research|Analyzed: Jan 3, 2026 08:53•
Published: Dec 31, 2025 03:43
•1 min read
•ArXivAnalysis
This paper introduces Recursive Language Models (RLMs) as a novel inference strategy to overcome the limitations of LLMs in handling long prompts. The core idea is to enable LLMs to recursively process and decompose long inputs, effectively extending their context window. The significance lies in the potential to dramatically improve performance on long-context tasks without requiring larger models or significantly higher costs. The results demonstrate substantial improvements over base LLMs and existing long-context methods.
Key Takeaways
- •RLMs are a novel inference strategy for handling long prompts in LLMs.
- •RLMs enable LLMs to recursively process and decompose long inputs.
- •RLMs significantly outperform base LLMs and existing long-context methods on various tasks.
- •RLMs can handle inputs far exceeding the model's context window.
- •RLMs offer comparable or cheaper cost per query.
Reference / Citation
View Original"RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds."