OpenAI’s GPT-6 Astra sparks AI agent adoption and memory crunch fears

  • OpenAI unveiled GPT-6 Astra, enabling users to run AI agents through ChatGPT.
  • Astra completed Pokemon FireRed in 18 hours and 12 minutes using screen analysis alone.
  • OpenAI’s next model is being trained with 400,000 graphics processing units, according to Jensen Huang.

OpenAI’s GPT-6 Astra is driving what observers call an "Astra moment," with AI agents that can complete multi-step tasks becoming accessible through ChatGPT. Nvidia Chief Executive Jensen Huang said on the 7th that Astra had reached artificial general intelligence, while the model has drawn attention for coding, media editing, office productivity and game-playing performance. Astra completed Pokemon FireRed in 18 hours and 12 minutes using only screen analysis, compared with 96 hours for GPT-5.6 Sol and more than 200 hours without completion for GPT-5.5. The model also earned a perfect 450 on the 2026 College Scholastic Ability Test, according to the developer community GitHub. Its recurrent depth approach repeatedly recomputes during inference, increasing total token use even as the price per token falls. Analysts expect this to expand demand for AI accelerators, HBM bandwidth and ultra-high-speed networking. OpenAI is already training its next model with 400,000 graphics processing units, four times the infrastructure used for Astra, while memory-market forecasts point to tightening supply through 2028.

本网站上的信息是使用AI生成的,我们无法保证其准确性。 请仅作为参考信息使用。