OpenAI releases GPT-6 Astra with advanced embodied intelligence capabilities, triggering industry-wide concern over data and compute barriers

OpenAI's GPT-6 Astra model demonstrates strong zero-shot performance in robotic tasks, achieving 95% success in block-placing tests versus 40% for Claude Fable 5.1. The model's capabilities stem from access to over ten million hours of embodied training data, highlighting a significant resource gap in the global embodied AI race.

Why it matters. The development underscores that data, compute, and time are critical barriers in embodied AI, potentially widening the gap between well-resourced AI labs and most robotics companies, especially in China where resources remain fragmented.