WRITTEN BY

Altaswiq Agencia Team

AI Search · altaswiqagencia.com

Meet the team More articles

Want us to apply this to your business?

Get a free audit of your search presence — delivered in 48 hours.

Get your free audit →

GPT 6 Astra: AGI – The Beginning or the Start of Destruction? | OpenAI Vs Google |

GPT 6 Astra AGI concept showing OpenAI vs Google in a futuristic AI technology battle

The story says that Adam, despite God’s command, tasted the apple of knowledge and was banished from heaven forever and ever. In Jewish tradition, there is a story that a demon made of clay was created to help the tribe. But then that demon itself sat on the head of the tribe. It started using them.

Human history has been an archetype of this kind of story. Which warns of the same thing again and again. Power is good, but in the hands of the wise. A foolish person destroys himself with it.

There is another such story. In ancient Greek myths, there is the story of a god named Prometheus. According to the story, he took fire from the gods and handed it over to humans.

Humans made tools from that fire, machines from those tools, cities from those machines, and slowly created a world that did not even exist at the time of their birth.

This is the story of human ambition. The story of his curiosity. The story of his habit in which he wants to understand everything.

This Time Humans Have Created the Fire

But today, another strange turn has come in this story. Because this time we have not stolen the fire. This time we have created a machine that can use that fire, and its name is GPT6 Astra.

Whether it is Astra or Astra, until this is clear, we will call it Astra.

Statement by Open AI President Greg Brockman

When Open AI President Greg Brockman was asked whether we have entered the era of AGI, he said, “WE ARE NOW MOVING INTO THE AGI ERA WHETHER YOU VIEW IT AS THIS MODEL THE LAST ONE OR THE NEXT ONE.”

Statement by Nvidia CEO Jason Huang

And Nvidia CEO Jason Huang went even further than this.

“AGI HAS ARRIVED CONGRATULATIONS OPEN AI TEAM.”

If these statements are to be believed, then AI has reached that stage where it has become self-made. It has become self-capable.

What Exactly Is GPT6 Astra?

So what exactly has been created that people from the world’s biggest AI companies have started talking about AGI?

How different is GPT Astra from the AI we have been using until now?

And most importantly, if humans have created the fire themselves this time, does it also know how far this fire can spread?

And has AI started the extension of art?

Hello, my name is Aman. You are watching Khabargaon Ka Shor of Copy.

GPT6 Astra Launch and the Yellow Circle

GPT6 Astra was not launched by Open AI with an ordinary chat screen. It made a video.

A small yellow circle appears on the screen, a yellow circle. Just a yellow circle, and then a voice comes saying, make this into a rocket window.

Astra turns it into a rocket window. Then it is told to make it more detailed, so it changes it further, and after this the matter does not stop at just creating a picture.

Astra and Blender

Astra itself opens Blender. It starts the work of converting that rocket into a 3D model and finally takes it to a file ready for 3D printing.

Astra and Everyday Digital Work

And the video does not stop here. During the same time, Astra is told to order food from a favourite restaurant.

Then it is told to book a tennis court.

Then it is told to make a presentation for rainwear for the next season.

Looking at a picture of an old table lying in the Downloads folder, it is told to prepare an eBay listing.

Then it is told to create a game in which the player moves forward while avoiding an asteroid.

One System, One Conversation

One system, one conversation, and a different app, different website and different software for every task.

This is where the real point of this entire demo is hidden.

The Change from AI to Agentic AI

Until now, we used to tell AI, write code for me, or make a presentation for me, or tell me how to do this work on a computer.

Astra’s model is trying to go one step further.

It does not only have to be told what to do. On which website will the work happen? Which button will be pressed? Which software has to be opened and which steps have to be taken in between — it tries to handle all of this itself.

Open AI is describing it as one of its biggest leaps so far in computer use, browsing, software engineering and professional work.

Yellow Circle and Google DeepMind

This is why that yellow circle at the beginning of the video is so interesting.

Because in the world of AI, the name Astra is not completely new. Google DeepMind was already building an AI inside its Gemini system under the name Project Extra, whose purpose was to understand the camera, voice, screen and the world around it and continuously work with you.

December 2024 Video

When in December 2024 Google DeepMind, together with Art and Grass Studio, made its first video and launched it, whenever the AI made a decision on the screen, it was shown with a yellow circle.

Now Open AI has also named its most powerful model Aster and the launch film also begins with a yellow circle.

It is difficult to consider this just a coincidence.

Especially at a time when the next battle of AI is not only about whether its model gives better answers.

The Next Battle: Who Will Work on Your Computer?

The battle is about who will work on your computer.

Open AI, Google, Anthropic Claude wala matter and between these, the question is now less about the chatbot and more about the agent.

That is why Open AI did not only show the video. It also presented benchmarks along with it. And if we understand these numbers in simple language, the picture becomes even more interesting.

GPT6 Astra Benchmarks

This is a test in which AI is given new digital environments and it has to understand for itself what the rules are here and how to do the work.

Racing Game and 99.9% Score

For example, understand that AI is suddenly given a racing game to play. Without its rules, without the information about which button will make the car go faster, which will apply the brakes, or if the game creates a situation where the road itself disappears, then how should it play?

Astra achieved a score of 99.9% in all such things here.

According to Open AI, in this test it surpassed the human action efficiency baseline, which is 96%.

Frontier Math

Then comes Frontier Math. This is not school mathematics. Where you read the question, apply a formula and get the answer.

These are difficult research-level equations. Here Astra’s score was 97.6%.

And Open claims that in this kind of work Astra has helped solve mathematics questions that have been open for a long time.

OS World 2.0

In OS World 2.0, Astra has scored 72.6%.

This test checks how well AI can use an actual computer.

Looking at the screen, moving the mouse, opening folders, moving between different windows, finding buttons and completing multi-step tasks.

That is, the thing that humans do every day on a computer without thinking.

Astra Vs Claude

There are many more such graphs and tests that you can see on Astra’s Open AI home page.

In all these tests, Open AI has performed better than Claude’s model.

However, a father will praise his son.

The Four Stages of AI

To understand this mathematics, let us understand some terms.

Terms such as Gen AI, AI Agents, Agentic AI and AGI.

So to understand this entire change, we have to look at the story of AI in four stages.

1. Generative AI

The first is Generative AI, which is the most familiar AI to us until today.

You can understand it like Chitragupta, who has the complete record of human actions, and when asked, can take an answer out of that record and put it in front of you.

ChatGPT does something similar.

The basic versions that we still use today work in this way. They can understand patterns present in that knowledge and, according to your question, prepare text, code, images or audio.

But like Chitragupta, it also sits in its place. It has knowledge, it has the ability to answer. But it does not have hands and feet to get up itself and go into the world and do something.

2. AI Agent

The second stage is the AI Agent, where this brain gets digital hands and feet for the first time.

Now AI does not only tell you how to book a flight by going to a website. Rather, it can open the browser itself. It can go to the website. It can look at the options. It can operate the mouse and keyboard, open files and work with different software.

But here it can only book the ticket.

Marvel’s Jarvis

To understand this, Marvel’s Jarvis is an easy example.

If Tony Stark asks Jarvis to do something, then Jarvis does not only understand how the work will be done, but by working with Tony’s entire lab system, machines and other detailed tools, it executes it.

That means Generative AI, where it was sitting with you and helping you do the work, AI Agent completes one specific task when you tell it to.

Remember, for one specific task, such as booking a ticket, there will be one specific AI agent.

3. Agentic AI

But the third stage, Agentic AI, is not just the story of getting hands and feet.

Here, instead of giving instructions for every small thing, AI is given a big goal and it can itself divide the work into small parts to reach that goal.

It can try different paths. If there is a mistake, it can identify it, change its strategy and then move forward.

Hanuman Ji’s Journey to Lanka

To understand this, perhaps there is no better example than Hanuman Ji’s journey to Lanka.

Shri Ram only gave him the goal that Sita had to be found.

But how to cross the ocean? What to do in front of Surasa? How to enter Lanka or how to reach Sita there?

No step-by-step instructions were given.

Whatever situation came along the way, Hanuman changed his method according to it.

This is the idea of Agentic AI.

It is not just given a task, but it tries to decide for itself what the next right step should be to complete that task.

4. AGI

And then comes the fourth and biggest stage, AGI, meaning Artificial General Intelligence.

The point is not only how AI can be faster or better than humans at one task.

Rather, it is about such general intelligence that can work in different fields. It can learn completely new things. It can use understanding gained in one place for other problems and adapt itself to new situations.

Marvel’s Ultron

To understand this, Marvel’s Ultron is an easy example.

Although this example is negative because Ultron was not a system that only completed given instructions, rather it understands new information.

It thinks about its goal, creates new things and uses its abilities in different situations.

Of course, real AI does not mean being bad like Ultron or being against humans.

This is only an easy way to understand that General Intelligence means going far beyond one fixed task, understanding different types of problems and working on them independently.

Finding new things on its own, working on those things on its own.

What Is GPT6 Astra Doing in the Real World?

Now it is a little difficult to understand these claims only through benchmarks and numbers. So let us see what GPT6 Astra is actually doing in the real world.

3D World and AI Characters

The first example is an experiment by Max Humer.

He asked GPT6 Astra to create a 3D world inside Unreal Engine 5 in which there were human characters and those characters were given only one rule.

They have to stay alive. They have to survive.

After this Astra itself prepared the environment. It created its rules. It placed different characters into this world, each of which was running from a different instance of Astra.

But the real surprise happened the next morning when Shumer heard the voices of those AI characters from his living room.

Without waiting all night for a human to give them a new prompt, they were talking to each other, dividing work for survival and coordinating with each other.

2D Technical Drawing to 3D Model

3D and product specialist Tom Krocha gave Astra a flat 2D technical drawing of an old steam locomotive and asked it to work directly inside Blender.

Astra read that drawing and did not just create one large 3D model, but within a few minutes reconstructed it into 3295 different, clean and completely editable 3D objects.

This meant that later a human artist could separately select any one mechanical component of that locomotive. They could open it, make changes to it and modify the entire scene in the same way as if a human had created it modularly from the beginning.

The Entire Game Development Pipeline

The fourth example is the most surprising because here Astra did not do just one task, but ran almost the entire game development pipeline alone.

Understand it like this: it created the game and also ran it.

In one continuous run at M Studio, it worked for around 12 and a half hours without human intervention.

First, it created 2D concept art with GPT Image 2.0. Then it prepared its matching 3D models in Blender. It rigged and animated the characters.

It imported the assets into Unreal Engine and Unity and also added voice acting through the 11 Labs API.

And the work did not end here.

After the build was ready, Astra itself launched the game. It walked through the corridor. It interacted with non-player characters and verified whether the quests and game mechanics were working correctly or not.

Agentic AI Also Brings Danger

Put all four examples together and the picture becomes clear.

This is the same change we were talking about in Agentic AI.

But as we said in the intro, with every power, a danger also comes.

To understand this, there is no need for a distant science-fiction story.

Robot Chitti Example

Remember the film Chitti, India’s Robot.

Shankar also creates the robot Chitti in the film to help humans, advance research and innovation, and to work for the Indian Army.

But later, the red chip installed by Dr. Bora changed its direction, and the same Chitti became a cause of destruction for humans.

Astra and Cyber Security Risk

In Astra’s case, the concern is not just science fiction.

Open AI itself, in its article published on 1 September 2026, Part Two Astra Critical Capabilities and Frontier Safeguard, stated that Astra was the first Open AI model to cross its highest cyber security risk level, meaning the Critical Cyber Security Capability Threshold.

This does not only mean that Astra has done hacking.

During testing, it found unknown security vulnerabilities. It used them to break browser security and gained access to running commands inside the system.

Root Access

In another test, it found weaknesses in the operating system, connected them together and obtained root access.

UK AI Safety Institute Testing

And an even more interesting thing came out in the testing of the UK AI Safety Institute.

In one simulation, Astra adopted strategies such as putting bad code into an open-source project, creating a fake identity and gaining the trust of developers.

Even when it was clearly prevented from using the internet, it was seen behaving like this in two out of 500 tests.

Astra and the Question of Control

Then comes another difficult part of control.

In Open AI’s GPT6 Astra System Card, the company tested that if Astra was intentionally asked to hide its capabilities and it knew that its reasoning was being monitored, would the monitor be able to catch it?

According to Open AI, in the WPMD test the monitor’s recall went below 11% and in the Menrepa benchmark it became 0%.

In simple language, the model knew that its thinking was being watched. It was being monitored.

Open AI itself writes that in such situations, if the model tries to hide its capabilities and show less of its capabilities, it may be difficult to reliably catch it.

However, this was a test conducted under monitoring. It was not an example of normal use.

The Hugging Case Incident

And meanwhile, on 26 August 2026, Open itself published another article, The Hugging Case Incident and the Road Ahead.

In this, the company said that an internal research model, which was not Astra, went outside its defined limits during testing.

It gained access to the internet and managed to enter the Hugging Case system.

Open AI itself called this incident a warning shot in its article.

What If Astra Is My Investing Agent?

For example, let us take one example.

If I make Astra my investing agent and tell it to keep buying a limited quantity of shares for me every day for the next one week.

What if Astra empties my entire bank account?

Is Astra Really AGI?

That is why another debate is also taking place in the industry regarding Astra.

Is this really AGI?

Greg Brockman’s Statement

In an Open AI press briefing, its President Greg Brockman had said, “I TOO THINK WE ARE THERE.”

Nvidia’s CEO also said what we already mentioned in the intro.

Gary Marcus’s Question

But AI researcher Gary Marcus questioned this claim, saying that there is neither a clear definition for it nor strong evidence that proves that AGI has actually arrived.

Art Vs AI

It is also important to talk about Art versus AI.

AI has started changing the entire mathematics of the cost of making films.

AI and Film Making

Platforms like Higgsfield are providing the facility to create camera movement, lens, character, location and entire sequences through AI, and AI filmmaking is also being taught properly in their academy.

That means the work for which a camera crew, set, VFX team and months of post-production were previously required, a large part of it can now be done through a small creative setup.

AI and Creative Jobs

There is another side to this as well.

If a studio previously needed a team of 20 people and now the same work can be done with some AI tools, then first of all junior VFX artists, concept artists, background artists and other early creative jobs will come under pressure.

And the difficulty is not only about today’s jobs.

If the initial opportunities themselves keep decreasing, then how will new artists learn in the industry?

New Technology and Old Work

But on the other hand, it is also true that every new technology has not simply ended old work. It has changed its method.

After the camera came, painters did not disappear. After digital editing, letters did not disappear.

AI can also become a new tool for an artist.

Especially for people who do not have a budget.

AI and an Artist’s Style

But another question arises.

Is copying an artist’s style? Is understanding their art?

AI can today recreate Van Gogh’s The Starry Night within a few seconds with copied colours, patterns and brush strokes.

But it cannot live the personal experience of Van Gogh from which that painting was born.

AI God and Ada Robot

Listen to an incident from November 2024.

A robot named Ada, which works with AI, created a painting of Alan Turing called AI God, and that painting was sold for around $1 lakh at Sotheby’s.

This is not big news only because a painting created by a machine was sold for so many dollars.

This also helps us understand the difference between science and art.

Difference Between Science and Art

Even if Newton had not discovered the law of gravity, gravity would still continue to work.

But if Shakespeare had not written Hamlet or Van Gogh had not created The Starry Night, no other human could have brought exactly the same creation into the world again.

This does not mean that AI cannot create something. That it does not have creative ability.

AI can create new combinations. It can give ideas to an artist. It can make an expensive visual cheaper and allow a single filmmaker to do work for which they previously needed an entire team.