8M tokens are the hook. 32k active tokens are the signal. RLM is not making the model hold the whole input. It treats long input as structured state the runtime can traverse. Runtime pages evidence…

Introducing a minimal training harness built on prime-rl and verifiers, so you can now train your own RLMs without sandboxes! All available in the `training/` folder in the RLM GitHub repo! We train…

Recursive LLMs process arbitrarily long prompts