ARC-AGI-4 Advance Teaser: Next-Gen Release Will Test Whether AI Can Independently Invent New Things

Beating AI News Brief: Nonprofit AI evaluation body ARC Prize has announced the upcoming ARC-AGI-4, its next-generation benchmark focused on testing "autonomous open innovation" — assessing whether AI can independently explore, generate new ideas, and even make inventions and discoveries. Specific test questions, scoring criteria, and release date have not been disclosed yet. ARC Prize noted that humans still hold a clear lead over AI in open innovation, a core capability driving scientific and technological advancement. The announcement also cites Dario Amodei’s recently published article on AI slowdown. The organization emphasized that open-source remains fundamental to AI progress, opposing industry efforts to reduce openness and concentrate cutting-edge AI in the hands of a small number of institutions in the name of coordinated slowdown. ARC-AGI has long been hailed as the world’s most challenging AGI benchmark. When ARC-AGI-3 launched, humans scored 100%, while leading AI models achieved just 0.51%. The latest GPT-6 Astra has made significant gains, but only scored 62.7% under the unified Standard harness; it jumped to 99.9% when paired with OpenAI’s proprietary Provider Adapter. ARC Prize is now set to advance to the next challenge: testing whether AI can independently drive innovation.