Agent: Cursor, Claude CodeLLM: GPT-4, Claude 3.5#LLM#inference#research#long-context#AI-agents
RLMs replace standard LLM completion calls with a recursive paradigm, enabling models to programmatically examine, decompose, and recursively call themselves over near-infinite contexts via a REPL sandbox. Built by MIT OASYS lab, backed by an arXiv paper.