Sam Altman drops bombshell in his podcast: Astra has reached human-level performance in operating computers.

Today, another exciting piece of news has arrived.

Sam Altman revealed in his latest podcast that Astra's ability to operate computers has reached human level!

Moreover, the launch of Fable 5.1 is just around the corner.

Some users found that Anthropic's latest model Fable 5.1 has quietly appeared in the API routing of Amazon AWS Bedrock!

It is very likely that Astra and Fable 5.1 will be released one after another around the same time.

The exciting moment is coming soon.

Astra's Computer Operation Ability Has Reached Human Level

At present, only a more than one-minute preview of this podcast has been released.

During the interview, the host raised this question: "Do you think AGI has arrived?"

Sam Altman replied: "It's close, at least extremely close... AGI is like a milestone, while superintelligence feels like an infinitely scalable monster. As for Astra, in terms of operating computers, it feels that it has fully reached human level."

In fact, recently this "human-level Astra" mentioned by Sam Altman has had a lot of actual test demos leaked quietly.

Last Sunday, a number of extremely shocking front-end demos of Astra were exposed.

Take the following test as an example.

With a simple prompt -- "Build a gorgeous sci-fi spaceship with an explorable internal structure" -- without any multi-turn dialogue iteration, the AI directly "one-shot" generated a complete 3D world.

As this KOL said, AGI may not arrive in the form of "chatbots becoming smarter", but appear quietly in the form of "creating a complete world with one sentence".

Moreover, OpenAI's ambition goes far beyond that.

Astra is not designed to compete with Fable for who is smarter at chatting, it is built for "persistent agents".

The details of Astra that have been disclosed so far are as follows.

Stunning 56K Tokens Inference: 3D Pagoda

Astra's latest output 3D voxel art pagoda work is extremely amazing.

At maximum inference intensity, consuming 56K inference tokens and taking 38 minutes, Astra recreated a 3D architecture from scratch that reaches an extremely high level in composition, color performance and overall rendering effect.

This reflects its extremely hardcore spatial imagination and code execution capability.

The "Never-Stop" Digital Worker

It is no longer a passive tool that "responds only when you ask a question". Astra can work continuously for several days or even weeks, it has long-term memory and can remember your corrections.

Multi-Agent Collaboration and Cross-Tool Operation

It can spawn multiple agents to jointly solve research-level mathematical proof problems.

What's even more impressive is that it can take over your computer desktop at "superhuman speed", collaborate across software, and even work together with other AI agents and humans.

This confirms what Sam Altman said in the podcast: "AI's ability to use computers has officially reached human level."

Fable 5.1 Appears in AWS Bedrock Routing

Moreover, there is new news today that the release of Fable 5.1 is not far away.

A developer announced a major discovery: Anthropic's latest model Fable 5.1 has quietly appeared in the API routing of Amazon AWS Bedrock!

When developers call the model identifier of Fable 5.1 in the system, the system originally returned a "400" error (the provided model identifier is invalid and is a made-up name); but now, it returns a "404" error (model not found).

This shows that AWS's underlying system has officially recognized and accepted the identity of Fable 5.1, and is just waiting for the model body to be uploaded.

This underlying anomaly is completely consistent with the performance a few days before the previous release of Opus 5.1.

In addition, in the past two weeks, many people have been included in the gray test, and all signs indicate that --

The countdown to the release of Fable 5.1 has already started, and it will be released in about 3 to 4 days.

If Anthropic cannot seize the ecological niche before Astra's release, Fable 5.1 is very likely to be completely outmatched.

In addition, some people tested Opus 5.1 in incognito mode, and the performance is also very good.

After Months of Actual Testing, Why Does Fable 5 Outperform GPT 5.6?

Furthermore, after months of actual testing by the well-known community Bridgemind recently, it has concluded that GPT 5.6 is outperformed by Fable 5.

Over the past few months, he has been using Fable 5 and GPT 5.6 Sol intensively every day, and his experience is as follows.

GPT 5.6 Sol is a powerful back-end machine, but it is defeated by "over-engineering".

In back-end processing, GPT 5.6 Sol performs better. It is extremely fast, has lower invocation costs, and performs better in data security.

Even in mathematical logic deduction, Sol has been able to significantly widen the gap with Fable and the previous generation of Opus.

Once the work enters an in-depth and complex stage, Sol is a more powerful back-end engine.

But it has an extremely fatal flaw: as long as the instruction is slightly loose, GPT 5.6 will commit the occupational disease of "over-engineering".

For a clearly simple function, it insists on building a whole set of huge and complex systems for you, and you need to restrict it extremely precisely.

As for Fable 5, it remains the de facto No. 1.

It wins in two core aspects: front-end UI design and "one-shot" capability.

When you throw a bunch of messy requirements to Fable 5, it often delivers a finished product that "feels already completed" on the first attempt.

Its extremely high first-pass success rate, coupled with its outstanding visual presentation and aesthetic taste, allows users to experience what real "smoothness" is.

This is why Fable 5 still ranks first in the hearts of developers -- it can not only get the work done, but also do it beautifully.

It is reported that Astra is now open for internal testing to some partners to collect feedback. If everything goes well, we will be able to experience this new AI species soon.

Next, who will take the lead, Astra or Fable 5.1?

References:

https://x.com/bridgemindai/status/2094406627663216733

https://www.youtube.com/watch?v=RX26fIVOg5E

https://x.com/pankajkumar_dev/status/2094399865971450286

This article is from the WeChat