We propose Recursive Language Models (RLMs), an inference strategy where language models can decompose and recursively interact with input context of unbounded length through REPL environments.

Hacker News is a community-driven platform for sharing and discussing technology news, startups, and programming-related topics. Through user submissions and comments, Hacker News offers insights into emerging technology trends, industry developments, and entrepreneurial ventures. Readers can participate in discussions, share their insights, and stay informed about the latest advancements in technology and innovation.

Hacker News

Researchers propose Recursive Language Models (RLMs), an inference strategy enabling language models to handle unbounded context lengths by recursively calling themselves through Python REPL environments. RLMs treat input context as a variable that models can programmatically interact with, decompose, and query recursively. Testing shows RLM using GPT-5-mini outperforms GPT-5 on difficult long-context benchmarks while being cheaper per query. The approach addresses context rot degradation in long conversations and scales to millions of tokens without requiring retrieval systems. Models naturally develop strategies like peeking, grepping, partitioning, and summarization when interacting with context. This represents a potential new paradigm for inference-time scaling alongside chain-of-thought reasoning and agent-based approaches.

Recursive Language Models

Prelude: Why is “long-context” research so unsatisfactory?

What We’re Thinking Now & for the Future.