What GPT-6 Astra Means for Robots, Agents and Drug Discovery: Three Researchers Weigh In

On October 23, 2026, Zhu Zheqing, founder of Pokee.ai; Xu Huazhe, founder of Pokebot and assistant professor at Tsinghua University; and Wang Sheng, assistant professor at the University of Washington, discussed in an interview how OpenAI's GPT-6 Astra, released in September 2026, is affecting personal agents, robotics, and AI-driven drug discovery.

Xu said Astra shows strong adaptability and self-iteration in robot tasks, with the ability to understand 3D space and quickly generate simulation environments from images. However, he noted that Astra is slow and still struggles with dynamic tasks requiring rapid responses, as well as with bimanual and dexterous manipulation.

Zhu said Astra delivers stable improvements over previous models in long-horizon tasks and iterative development, with better context awareness and fewer hallucinations. But he added that it remains relatively slow and hallucinations still occur in very long tasks, so the issue is not fundamentally solved.

Wang highlighted that Astra's Computer Use and long-horizon capabilities are practical in biomedical work, such as operating legacy scientific software and running tasks lasting eight to 24 hours. Still, he remained cautious about AI in drug discovery, citing slow real-world feedback loops and noting that AI-designed experimental plans are only about 30 percent complete, which remains a major bottleneck.