Abstract Proposes Recursive Language Models (RLMs), a general inference paradigm that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt.

Key Breakthroughs

• Unbounded Context: Successfully processes inputs up to two orders of magnitude beyond model context windows.
• RLM-Qwen3-8B: The first natively recursive model, outperforming its base model by 28.3% on average and approaching GPT-5 quality on long-context tasks.
• Programmable Inference: Treats long prompts as external environments for programmatic traversal.

RSI Impact (yanhua.ai) Eliminates the "Context Bottleneck" in recursive self-improvement. Confirms that natively recursive architectures (like RLM-Qwen3) are superior to simple agentic scaffolds for long-horizon evolution.

Links View on ArXiv | GitHub Repository