A reminder that I’m on vacation this week, but the flood of impressions people have posted online over the past few days as they gained access to OpenAI’s GPT-6 Astra, seems like an important moment to briefly cover.

GPT-6 is likely to go down as a big milestone in AI development.

OpenAI made Astra a full point release to version 6 for good reason. After reading hundreds of posts from credible, reputable developers and AI experts I follow, it’s clear the model is strong at coding and other typical model work.

“Putting benchmarks aside, I deal in financial data and can tell you Astra devastates Fable,” said Houman Dehesh, founder of Helion Capital, adding that he believes the model must have been trained on a large amount of enterprise data.

But what’s really impressing people is Astra’s ability to handle vision and computer use. OpenAI made computer use the centerpiece of its Astra marketing video, which is worth watching here.

“Computer use is a particularly important part of what’s new. I think that it has crossed the qualitative threshold. And the fact that it is now, you’ve gone from this slow, painful to watch thing to something that’s actually very accurate, efficient, and reliable. So Astra can really do anything a human can do with a computer without an API.” — OpenAI cofounder Greg Brockman said in an Astra press briefing.

The last time I tried computer use, where the AI model takes control of the screen to navigate the web and operate software on the user’s behalf, was a year ago, with the release of ChatGPT Atlas. Back then, it was too slow and clumsy. I was eager to see how far OpenAI has come since.

I tested Astra over the weekend. It handled the basic financial spreadsheet tasks I need with ease, but I was more interested in whether it lived up to the computer use hype.

I had never used the 3D program Blender before, so I downloaded it onto a spare Mac laptop and asked Astra to build me a detailed model of the Space Shuttle. I watched as it researched the topic online, wrote its own scripts, and started operating Blender on its own.

Watching it take over the program felt like magic. A glimpse of the future. Within 10 minutes, I had a beautiful, detailed 3D model of the shuttle that I could zoom in and out of and rotate.

With the release of GPT-6 Astra, computer tool use AI has finally become real. I am confident now that AI will soon be able to perform actual software knowledge work tasks better and faster than most humans, and its competence is only going to accelerate rapidly from here.

Computer use will still requires human taste and steering, but it should enable incredible innovation, rapid iteration and experimentation, and real gains in efficiency and productivity.

I also predict the use of AI models for computer use and knowledge management tasks will follow the same trajectory as agentic coding.

Here is a clue of the future path. On Sunday, OpenAI said in a blog post that the median researcher at the company used a negligible amount of coding agents at the start of this year, but by mid-August they had integrated agents into their workflow to the tune of more than $600 worth of tokens per day.

“At the start of this year, the median researcher ranked by agent usage at OpenAI was using coding agents only in modest amounts. By mid-August, the median researcher was integrating agents daily into their work, using more than $600 per day of inference at API prices. The 90th percentile user in our research organization now uses more than $7,000 of tokens per day.” - OpenAI

As computer tool use rises, AI compute demand will explode just as it did when agentic coding took off, when reasoning models ramped, and when AI chatbots were first released.

Computer use is going to be the FOURTH exponential demand wave in FOUR years (chatbots → reasoning → agentic coding → agentic computer tool use). As AI compute demand goes exponential again from this fourth wave, demand for AI semiconductors is likely to surge with it (CPUs, HBM memory, and GPUs) just like the last three times.

I’ll say it again. We’re still in the early innings of the AI mega-trend. Yes, we’re just getting started.

Take a look at some of the other Astra demonstrations online that showcase its capabilities:

-3D house architecture modeling

-3D interactive atlas of human ankle

-3D human anatomy

-Tendon surgery demo

-3D forest home

-Compose piano melody and 3D pianist