Nvidia CEO Jensen Huang has declared that “AGI has arrived,” crediting OpenAI’s newly released GPT-6 Astra model — a claim OpenAI itself has not made in those exact terms, and one that has drawn both excitement and skepticism across the tech industry.
What Happened?
On September 6, 2026, Huang posted on X: “GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next.” OpenAI launched GPT-6 Astra in early September 2026, describing it as its most intelligent and aligned model to date, with the company’s president Greg Brockman calling it a step into the “AGI era.”
Latest Update
According to OpenAI, Astra posted state-of-the-art results on demanding benchmarks, including 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench, alongside strong performance on TerminalBench-4.0 and HealthBench Pro. The model is rolling out in phases to ChatGPT Plus, Pro, Business and Enterprise subscribers, as well as through the OpenAI API and cloud partners including Amazon Web Services, with wider access expanding gradually rather than to all users at once.
Why Is GPT-6 Astra Trending?
The story is trending because the “AGI has arrived” claim came from Nvidia’s CEO rather than OpenAI itself, prompting debate over whether the declaration reflects genuine technical achievement or serves as a promotional boost for Nvidia’s chip business, given that Nvidia supplies the hardware Astra was trained on.
Key Details About Astra
- Model name: GPT-6 Astra, from OpenAI
- Training hardware: Reportedly over 100,000 Nvidia Grace Blackwell NVLink72 systems
- Claimed strengths: Software engineering, science, mathematics, cybersecurity, agentic/computer-use tasks
- Benchmark scores cited by OpenAI: 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, 100% on ExploitBench
- Availability: Rolling out to ChatGPT Plus, Pro, Business and Enterprise users, plus the OpenAI API
- Notable caveat: OpenAI has said additional safety and alignment work continues alongside the wider rollout
The Road From ChatGPT to Astra
Huang framed the announcement as rapid progress “from ChatGPT to o1 to Astra in 4 years,” referencing OpenAI’s earlier reasoning-focused “o1” model. This isn’t the first time Huang has made an AGI claim — he made a similar statement on the Lex Fridman Podcast in March 2026, roughly six months before Astra’s release, saying “I think it’s now. I think we’ve achieved AGI.” Critics point out that AGI (artificial general intelligence) has no single agreed-upon technical definition, and that Nvidia has a direct commercial interest in promoting demand for the hardware used to train frontier models like Astra.
What Happens Next?
OpenAI has not echoed Huang’s specific “AGI has arrived” framing, and independent evaluation of Astra’s real-world capabilities will likely continue as wider access rolls out to more users. Huang has also said a further 400,000 Nvidia GPUs are “coming online next,” suggesting continued heavy investment in training future models.
FAQs
Q: What is GPT-6 Astra?
It’s OpenAI’s newest AI model, described by the company as its most intelligent and aligned model to date, with strong benchmark performance in software engineering, science, and cybersecurity.
Q: Did OpenAI say AGI has arrived?
Not directly in those words. The “AGI has arrived” claim came from Nvidia CEO Jensen Huang; OpenAI’s president described it as entering the “AGI era” without a formal AGI declaration.
Q: How many GPUs were used to train GPT-6 Astra?
Huang said the model was trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems.
Q: Can I use GPT-6 Astra right now?
It’s rolling out in phases to ChatGPT Plus, Pro, Business and Enterprise subscribers, as well as via the OpenAI API, rather than to all users simultaneously.
Q: Why are some people skeptical of the AGI claim?
Because Nvidia sells the hardware used to train Astra, and AGI has no universally agreed technical definition, critics say the declaration may be partly promotional.
