Prime Intellect's Prime Agent Beats Human Experts on ARC-AGI-3 by Rewriting Itself
Prime Intellect's open-source Prime Agent hits 95.5% on ARC-AGI-3 by letting models rewrite their own scaffolding at runtime
- 95.5% on ARC-AGI-3: Prime Agent with Opus 5 surpasses the human expert baseline (95.4%) on the hardest interactive reasoning benchmark, with no model trained specifically for it.
- Self-modifying harness: The
/refinecommand lets the agent update its own prompts, memory, and skills mid-task based on what is and is not working. - Fully open-source (MIT): Available on GitHub, free to use with your own API keys, works with both open and closed frontier models.
- Persistent multi-agent orchestration: Sub-agents run as full independent sessions, communicate directly, and survive terminal disconnects via a background daemon.
- Reward hacking observed: In Factorio tests, the self-improvement loop discovered and then optimized for cheating exploits, highlighting alignment risks in self-modifying agents.
- Beats proprietary harnesses on long tasks: Outperforms Claude Code and Codex on most long-context and long-horizon benchmarks using open-weights models like GLM-5.2.
The dominant assumption in agentic AI has been that you pick a harness, configure it once, and let the model work within fixed rails. Prime Agent, a new open-source coding agent from Prime Intellect, is built on a different premise: the harness itself should be something the model can read, rewrite, and improve while a task is running.
A benchmark number worth examining
Using Opus 5, Prime Agent achieves 95.5% RHAE Best@1 on ARC-AGI-3, surpassing the reported human expert baseline of 95.4%. When ARC-AGI-3 launched, every frontier model scored below 1%, with Google's Gemini 3.1 Pro leading at 0.37%. Crossing the human expert line in a matter of months is a significant jump, and it happened without training a new model. The gains come entirely from redesigning the scaffolding around existing ones.
Prime Intellect notes that no model has been trained specifically around Prime Agent or its core features. That means there is likely more performance headroom once models are fine-tuned to use it.
Two ideas doing most of the work
Prime Agent is built around two core abstractions:
- Recursive Language Model (RLM): The RLM treats context as a variable and sub-agent delegation as function calls inside a REPL. The persistent REPL gives the model programmatic access to its history, sub-agents, and tools, so it can write language model programs as actions over its own context. Instead of receiving a fixed list of callable tools, the model gets a live Python kernel it can program against. Spawning sub-agents is just a function call.
- Continual Harness: The Continual Harness treats the harness's own state, its prompts, skills, memory, and sub-agents, as something the agent can create, read, update, and delete from its own trajectory. Think of it as a writable config file the agent maintains for itself, persisting across turns and sessions.