OpenAI has introduced its latest model, GPT-6 Astra, which the company claims demonstrates AGI-like capabilities. The model was trained using 100,000 NVIDIA Grace Blackwell NVLink72 GPUs, with plans to deploy an additional 400,000 units in the near future.
According to the Artificial Analysis Intelligence Index, Astra is currently more cost-effective per task than Anthropic's Fable 5.1, though it remains 50% more expensive than the GPT-5.6 Sol model. Despite its performance, Fable 5.1 currently holds the top position on the index.
OpenAI researcher Jakub Pachocki emphasized that the company is focusing on developing AI as a research tool, with future goals including scientific progress and personalized AGI assistants. He noted that AI currently simulates human behavior rather than recreating it.
Amid growing competition from models like Deepseek V4, Kimi K3, Gemini Flash 3.8, and Muse Spark 1.3, OpenAI is advocating for international government oversight and a potential slowdown in AI development. The company suggests that legislation may be necessary to ensure human control.
The release of Astra occurs as OpenAI transitions toward a for-profit structure, with industry observers noting a potential upcoming IPO. Calls for regulatory intervention are viewed by some as a strategic move to manage competitive pressures in the rapidly evolving AI market.
