GPT-6 Astra is OpenAI’s first model at Critical under its Preparedness Framework; Nvidia’s CEO called it AGI on social media.
Technology • AI • Governance
OpenAI released GPT-6 Astra with a Sep. 3 safety overview calling it the company’s first model to reach the Critical level of cybersecurity capability under its Preparedness Framework. That designation is OpenAI’s own language, not a government rating. Standard API list pricing is $10 per million input tokens and $50 per million output tokens for short context.
On ARC-AGI-3, Astra scored 62.7% under the Standard harness and 99.9% under a Provider Adapter harness; ARC Prize says saturating the benchmark is not proof of AGI. Nvidia CEO Jensen Huang posted Sep. 6 that Astra was trained on roughly 100K+ Grace Blackwell systems and that “AGI has arrived,” with 400K GPUs “coming online next.” Those lines are Huang claims, not established fact.
Sources: OpenAI safety overview and Path to Astra; ARC Prize; Huang X post via Business Insider/TNW.
Cedar Global Markets
