Nvidia CEO Jensen Huang declares AGI has arrived with OpenAI's GPT-6 Astra

September 7, 2026 at 07:56 PM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Nvidia CEO Jensen Huang declared on social media that artificial general intelligence has arrived with OpenAI's GPT-6 Astra, trained on over 100,000 Nvidia GPUs. OpenAI president Greg Brockman welcomed the AGI era, while the company highlighted Astra's performance on software engineering, science, and math benchmarks. Huang's announcement also promoted Nvidia hardware, calling for 400,000 GPUs to be brought online.

Key Facts

OpenAI's GPT-6 Astra was trained on a cluster of over 100,000 Nvidia GPUs, according to Nvidia CEO Jensen Huang.
OpenAI president Greg Brockman said this was the first training run to use more than 100,000 GPUs.
Huang stated that 400,000 GPUs are planned to be brought online in anticipation of future models.
Wccftech estimates the GPT-6 Astra training run cost between $500 million and $1 billion, assuming $2.50–$3.50 per GPU hour.
OpenAI says GPT-6 Astra is state-of-the-art on benchmarks including FrontierMath Tier 4, ARC-AGI 3, and TerminalBench-4.0.

The AGI Claim

Nvidia CEO Jensen Huang posted on X that 'AGI has arrived' with the release of OpenAI's GPT-6 Astra. OpenAI president Greg Brockman echoed the sentiment, saying 'welcome to the AGI era' during the model's debut. Brockman acknowledged that AGI is a 'blurry' concept, noting that people have their own definitions. He added that 'somewhere in there, I think we're going to cross most people's AGI threshold,' referring to recent or upcoming models.

Training Scale and Cost

GPT-6 Astra was trained on '~100K+ Nvidia Grace Blackwell NVLink72' server racks, according to Huang. Wccftech estimates the training run cost between $500 million and $1 billion, based on $2.50–$3.50 per GPU hour over 90–120 days. The training likely took place at OpenAI's Stargate data center in Abilene, Texas, according to Wccftech. Huang said plans are already in place to bring 400,000 GPUs online for future training runs.

Benchmark Performance

OpenAI describes GPT-6 Astra as 'state-of-the-art' on benchmarks including FrontierMath Tier 4, ARC-AGI 3, and TerminalBench-4.0. The model shows strong aptitude in software engineering, science, mathematics, and professional workflows, according to OpenAI. Anthropic's Fable 5.1 still outperforms GPT-6 Astra on SWE-Bench Pro and coding benchmarks, according to Wccftech.