If your work is computer-related, you might as well take up farming—I'm hearing these kinds of things online. I don't understand why there is an atmosphere of fear with every AI model launch. The FOMO from Fable 5 is repeating itself after the launch of ChatGPT-6 Astra. Everyone is saying we've achieved AGI. Bernie Sanders is out there talking about permanently banning AI. Man, did you forget what happened with Fable 5? Just watch this video a month from now and you'll understand what I'm saying.
The same AI hype cycle keeps repeating
There's a cycle that repeats with every model launch. In the next minute, you'll understand what all this hype is about, what this model is genuinely good at, and whether we've actually achieved AGI.
First of all, it's not like it launched today and everyone will start using it tomorrow. Initially, it will roll out to companies in OpenAI's cyber defender program, Daybreak. The rest of the Plus, Pro, Business, and Enterprise users will get it in the coming days. In fact, its absolute best version, Astra Pro, will only be available to the top paid tiers.
What GPT-6 Astra is actually good at
Secondly, you can guess how expensive it is by the fact that DeepSeek's flagship model V4 Pro costs around $1 to $2 per million output tokens, while GPT-6 Astra costs $50. Now, in coding, this model is still behind Anthropic's model. But the climate of fear exists because it scored 99.9% on ARC-AGI 3.
Think of ARC-AGI like this: the AI is placed in a completely new environment without being told the rules or the goals. It has to figure them out itself, take actions, and then observe the consequences to understand that environment. This means that while earlier AI would only do what it had seen before, now if you give it new software or a task, it will largely figure it out on its own.
It is also exceptionally good at computer use—it's the best model to date for it. If your work has a fixed workflow, this model can operate your computer and perform that task quite well. As for its flaws, those will only become clear over time. One more area where it excels is 3D; it can use software like Blender very effectively to create something meaningful.
Have we actually achieved AGI?
Now, let's talk about AGI. I said it a month ago and I'll say it again today: before real AGI arrives, ten fake AGIs will come and die out. Whenever someone mentions AGI, a few questions come to mind: Is this model robust in its understanding of the physical world? Does it continually learn from its previous experiences? Does it reliably adapt to new situations without needing retraining? Can it reliably complete long-horizon tasks without breaking down, or does it collapse after reaching a certain level of complexity?
There's no doubt that this Astra model has made very significant progress. But just because it saturated one or two benchmarks, calling it AGI based on that would be very premature. What will happen is that once it gets into people's hands, they will discover its flaws with open-ended, real-world problems.
What is left between human and artificial intelligence?
It's only been a few hours since its launch. Following this launch, a question popped into my head: what is the major difference left between human intelligence and artificial intelligence?
The useful takeawayBefore real AGI arrives, ten fake AGIs will come and die out. Just watch this a month from now and you'll understand what I'm saying.
Need help putting this into practice? Explore our ai content services.