Full Breakdown
GPT-6 Astra: OpenAI’s Artificial General Intelligence (AGI) Claim and the Industry’s Response
9/7/2026, 7:53:03 AM
Core Event – Launch of GPT-6 Astra and the AGI Assertion
OpenAI released GPT-6 Astra, branding it the world’s “most intelligent and aligned model.” In a media call the company’s president declared, “Welcome to the AGI era,” suggesting the model could mark the creation of artificial general intelligence. The launch was announced on September 3, and the model became available to Pro, Enterprise and Business Premium users within days.
Background & Context – From ChatGPT to a Massive GPU Super-cluster
Astra follows OpenAI’s recent releases—ChatGPT, the “o1” system, and now Astra—trained on Nvidia GPUs. Nvidia’s chief executive noted, “From ChatGPT to o1 to Astra in 4 years,” highlighting the hardware partnership. OpenAI’s internal blog disclosed that the training run used more than 100 000 GPUs in a Texas-based “Stargate” super-computer, the largest training task in the company’s history.
Data & Statistics – Scale, Revenue and Compute Consumption
| Metric | Figure | Source |
|---|---|---|
| Nvidia quarterly revenue (August) | $96.2 billion (double YoY) | Nvidia report |
| Data-center revenue (AI chips) | $89 billion | Nvidia report |
| GPUs slated for Nvidia’s fleet | 400 000 units | Jensen Huang’s X post |
| Median OpenAI researcher daily inference cost | ? $600 | OpenAI internal data |
| Top 10 % researcher daily inference cost | > $7 000 | OpenAI internal data |
| AI-to-human work-day ratio (mid-2026) | 3.1 AI days per human day | OpenAI internal data |
| Agents run concurrently by >70 % of researchers | 4–5 agents each | OpenAI internal data |
These figures illustrate the compute demand that Astra required for training and now generates during inference.
Official Statements & Responses
- Jensen Huang (Nvidia CEO) – Congratulated OpenAI on X, asserted that “AGI has arrived,” and announced that 400 000 GPUs will be added to Nvidia’s infrastructure.
- Greg Brockman (OpenAI President) – Called Astra a “generational leap” and welcomed the “AGI era,” while noting the lack of a universal AGI testing standard.
- Sam Altman (OpenAI CEO) – Previously called AGI a “very poorly defined term,” later suggesting Astra could be the point historians identify as AGI emergence.
- OpenAI Security Team – Classified Astra at the “Critical” level of its Preparedness Framework, restricting offensive-security capabilities.
Criticism & Opposition – Questioning the AGI Declaration
AI researcher Gary Marcus criticized Huang’s proclamation, arguing that “Huang gave no evidence and no definitions,” and posted a ten-point AGI definition, noting Astra meets only one or two criteria.
Conflicting Reports & Gaps – Disagreement Over What Constitutes AGI
OpenAI frames Astra as a practical realization of AGI, while Altman calls the term “poorly defined.” Analysts, including Marcus, contend the model falls short of established AGI benchmarks. No universally accepted metric exists, and the security classification shows Astra’s offensive capabilities remain limited, indicating a gap between performance and the broader definition of “general intelligence.”
What’s Next – Hardware Expansion and the Automated AI Research Intern
- Nvidia plans to bring 400 000 GPUs online soon to sustain compute growth.
- OpenAI announced completion of its “Automated AI Research Intern” as of September 2026 and targets an “Automated AI Researcher” by March 2028.
The launch of GPT-6 Astra couples a bold AGI claim with unprecedented hardware investment, while the community debates whether the model truly meets the standards of general intelligence.
