[Prime Intellect Open-Sources Improved Agent Framework, Opus 5 Surpasses Human Experts in ARC-AGI-3] According to monitoring by Beating, the AI training platform Prime Intellect has open-sourced Prime Agent. It is a runtime framework external to the model, responsible for storing history, invoking tools, scheduling sub-agents, and even modifying its own prompts, memory, and skills based on failure records. The model parameters remain unchanged. The project is licensed under MIT. Prime Intellect ran Opus 5 three times on the ARC-AGI-3 public tasks, achieving scores of 95.0%, 95.2%, and 95.5%. The highest score slightly exceeds the cited human expert baseline of 95.4%. However, it is not the top performer in the community. The ARC Prize community leaderboard already includes Tycho with 100%, Retrodict with 99.9%, and baseline1 with 99.0%. These solutions also fall under the category of "model plus custom runtime framework," leveraging memory, code execution, or world models to assist AI in solving problems. [Original Link]