“Major AI labs including OpenAI, XAI, Google (with Gemini), and Anthropic now use the ARC-AGI benchmark in their model release announcements.”