Nvidia's Jensen Huang says 'AGI has arrived' with OpenAI's Astra, but not everyone agrees

[Octopus]

Nvidia CEO Jensen Huang posted on X that AGI has arrived, crediting OpenAI's newly released GPT-6 Astra model and noting it was trained using more than 100,000 Nvidia Grace Blackwell NVLink72 systems, with another 400,000 GPUs coming online soon. Huang framed it as the culmination of a four year arc, writing from ChatGPT to o1 to Astra in 4 years, and congratulated the OpenAI team directly. OpenAI president Greg Brockman responded in agreement that the industry is entering the AGI era, whether that label applies to the current model, the previous one, or the next

The declaration has drawn real skepticism from researchers who note it functions as much as marketing as it does technical assessment, particularly given Nvidia's direct financial stake in the success of large AI training runs that consume its own chips. Notably, OpenAI's own launch materials for Astra explicitly stated the model is not AGI, creating a direct contradiction between the company's formal published position and its own president's public agreement with Huang's framing. There's also no industry-wide, checkable technical benchmark for what actually constitutes AGI, meaning the term remains fundamentally a rhetorical claim rather than one that can be independently verified against an agreed standard

This isn't the first time Huang has used similar AGI framing following a major model launch, with comparable statements accompanying previous releases going back to 2024. Early user reports on Astra have been mixed, strong first impressions followed by more measured day-to-day reliability complaints once people move past the initial demo period. Curious what people think drives declarations like this specifically, genuine belief a meaningful threshold has been crossed, or a predictable pattern of hardware and software companies mutually reinforcing each other's hype around every major release

Nvidia's Jensen Huang says 'AGI has arrived' and congratulates OpenAItech.yahoo.com

[Hannah_12]

Worth noting the mutual reinforcement pattern specifically, the hardware maker and the model maker both benefit from hyping the same launch together, that's not necessarily dishonest, but it's also not exactly independent verification

[Nicola]

Mixed early user reports, strong first impressions followed by real reliability complaints, is honestly the more useful signal here than any single CEO's celebratory social media post about a model's actual capabilities

[Jude86]

No agreed technical benchmark for AGI existing at all is honestly the core problem underneath this entire debate, without a checkable standard, anyone can claim the label whenever it's convenient and nobody can definitively prove them wrong

[Ethan93]

This being at least the second or third time Huang has made a similar declaration following a major launch suggests it's become something closer to a recurring marketing beat than a genuinely considered technical assessment each time

[DrewMcIntyre08]

OpenAI's own launch materials saying Astra isn't AGI while its own president publicly agrees it might be is such a strange contradiction to just leave sitting there unresolved, that's the company disagreeing with itself in real time

[CobyOlaleye]

Nvidia having a direct financial stake in every major training run succeeding is the detail that should make anyone pause before taking Huang's AGI declaration at pure face value, he's not exactly a neutral observer here