By: Synzhiyuan
Today, there’s even more exciting news.
In the latest podcast, Altman revealed that Astra’s ability to operate a computer has already reached human level!
And Fable 5.1 isn’t far off either.
It was discovered that Anthropic’s latest model, Fable 5.1, has already quietly appeared in the API routes of Amazon AWS Bedrock!
It’s very likely that Astra and Fable 5.1 may be released around the same time, back-to-back.
Exciting moments are just around the corner.
Astra operating a computer has reached human level
So far, this podcast has only released a teaser of more than a minute.
In the interview, the host asks: “Do you think AGI has already arrived?”
Altman answers: “We’re close—at least we’re very close... AGI feels like a milestone, while superintelligence feels like a monster that can expand infinitely. As for Astra, when it comes to operating a computer, it already feels fully at the level of humans.”
Actually, recently, a lot of real-world demo evidence about the “Astra at human level” mentioned by Altman has quietly leaked out.
Last Sunday, Astra already revealed a bunch of extremely impressive front-end demo showcases.
For example, the following test.
With a single simple prompt—“Build a magnificent sci-fi spaceship, with an internal structure that can be explored”—without any multi-round dialogue iteration, the AI directly “one-shot” a complete 3D world.
As this big-name figure said, AGI likely won’t arrive in the way of “a chatbot getting smarter.” It may instead quietly emerge as “one sentence that creates an entire world.”
And OpenAI’s ambition goes far beyond that.
Astra isn’t meant to compete with Fable in who can chat more intelligently; it’s built for “continuous agents.”
The Astra details disclosed so far are as follows.
Astonishing 56K tokens for reasoning: a 3D pagoda
Astra’s latest output—this 3D voxel art pagoda—looks truly astonishing.
At maximum reasoning intensity, Astra consumed 56K reasoning tokens over 38 minutes, and somehow managed to recreate a 3D architectural masterpiece at an exceptionally high standard in composition, color rendering, and overall visual output.
It reflects extremely hardcore spatial imagination and code execution capability.
“Never-stop” digital workers
It’s no longer a passive tool that goes “you ask, it answers.” Astra can work continuously for days or even weeks. It has long-term memory and can remember your corrections.
Multi-agent collaboration and cross-tool operations
It can split into multiple intelligent agents, coordinating to tackle research-level mathematical proof problems.
Even more terrifying is that it can take over your desktop “at superman-like speed,” collaborate across software, and even work together with other AI agents and human teams.
This confirms what Altman said on a podcast: “The capability of AI to use computers has officially reached the level of humans.”
Fable 5.1 shows up in the AWS Bedrock API routing
And today, there’s more news: Fable 5.1 isn’t far off.
Developers have published a major discovery: Anthropic’s latest model, Fable 5.1, has quietly appeared in the API routes of Amazon AWS Bedrock!
When developers call the Fable 5.1 model identifier in the system, it originally returned a “400” error (the provided model identifier was invalid—made-up nonsense). But now it returns a “404” error (model not found).
This shows that AWS’s underlying system has officially identified and accepted Fable 5.1’s identity—it’s only waiting for the model itself to be uploaded.
This underlying shift matches the exact same behavior as what happened just a few days before the release of Opus 5.1 earlier.
And over the past two weeks, many people have been spotted in gray tests. All signs point to the fact that—
The release of Fable 5.1 is already in the countdown—roughly within the next 3 to 4 days.
If Anthropic can’t secure its ecosystem foothold before Astra is released, Fable 5.1 will very likely be an instant kill.
Besides that, someone has also tested Opus 5.1 in stealth mode, and the results are also quite impressive.
After months of real-world testing, what makes Fable 5 outshine GPT 5.6?
And recently, after months of testing, the well-known community Bridgemind came to the conclusion: GPT 5.6 gets beaten by Fable 5.
Over the past few months, he has been mixing Fable 5 and GPT 5.6 Sol at high intensity every day—here’s his firsthand experience.
GPT 5.6 Sol is a powerful backend machine, yet it loses to “over-engineering.”
In backend processing, GPT 5.6 Sol performs even better. It’s extremely fast, cheaper to call, and does a better job on data security.
Even in mathematical logic reasoning, Sol has already managed to pull a significant gap away from Fable and its predecessor, Opus.
Once the work enters a deep and complex stage, Sol is the stronger backend engine.
But it has one extremely fatal flaw: as soon as the instruction is even slightly loose, GPT 5.6 falls into the professional illness of “over-engineering.”
It’s a simple feature, yet it insists on building an entire huge, complex system for you—you need to constrain it extremely precisely.
And Fable 5—it's still the de facto No.1.
It wins in two soul-stabbing places: front-end UI design, and its “one-shot” capability.
You throw a bunch of messy, chaotic requirements at Fable 5. It often produces a finished product that already feels “basically done” on the very first attempt.
Its extremely high launch-time hit rate, combined with outstanding visual presentation and an impeccable aesthetic sense, gives users what true “silky-smooth” experience feels like.
That’s why Fable 5 still ranks #1 in developers’ hearts—it's not only capable of getting the job done, but it does it beautifully.
There are reports that Astra has opened an internal beta collecting feedback to some partners. If everything goes smoothly, we’ll be able to experience this new species soon.
Next, who will make the first breakthrough—Astra, or Fable 5.1?