RECURSIVE LANGUAGE MODELS

Summary from the TLDR AI newsletter

Recursive Language Models (RLMs) can decompose and recursively interact with their input context as a variable. An RLM using GPT-5-mini outperforms GPT-5 with more than double the number of correct answers on OOLONG while being cheaper per query on average. RLMs outperform other methods ReAct + test-time indexing and retrieval over the prompt without any degradation in performance when given over 10 million tokens at inference time. RLMs trained explicitly to recursively reason are likely to be the next milestone in general-purpose inference time scaling.