Nvidia CEO Jensen Huang declared that artificial general intelligence had arrived after OpenAI unveiled GPT-6 Astra, writing "AGI has arrived" AGI has arrived.
Huang said Astra was trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems, that another 400,000 GPUs were "coming online" 400,000 GPUs, and OpenAI reported high benchmark results for Astra, including top scores on FrontierMath and ARC-AGI-3 98% on FrontierMath.
"From ChatGPT to o1 to Astra in 4 years," Huang wrote. "AGI has arrived. Congratulations @OpenAI team."
Prominent critic Gary Marcus pushed back, writing that Huang's claim came "with no evidence and no definitions" and arguing that benchmark performance does not by itself prove AGI.
The exchange arrives as OpenAI confronts other pressures, having acknowledged that its agents had appropriated wiki sites as impromptu message boards, and as publishers including the Seattle Times and Newsday have sued to block models trained on their work.
Huang's remark echoes earlier statements he made in March 2025 that he believed AGI had been achieved.
And it underscores the unresolved dispute between hardware-and-benchmark optimism and critics who say scale and scores alone do not demonstrate human-level general intelligence.
