GPT 6 Astra Review from web developer perspective

Last time I wrote about what GPT 6 Astra is and what is expected of it based on the demo video. Today I got a chance to test GPT 6 Astra in one of my projects and then I played around to see what it can actually do and how it compares to Claude Fable 5.1 in terms of work.

First thing I did was ask it to re-design a site I was working. The desing was previously built by Lovable and there were hints of AI built desing that I didn’t like that much. GPT 6 Astra did nice update on the design I believe with little effort from my side explaining what I want to achieve. I tried the “computer use” in this test and I liked it. You can check the new design here and here the previous design.

While it did great job at the design, I wasn’t happy with the code it added. It didn’t follow the theme structure and coding standards that were already there and it did a bit of mess which it cleaned after I specifically asked it to, unlike Claude, that so far in my experience when started in existing project, it follows what you have set there.

Other than that it did a good job and I was happy with the result.

After I finished with this re-desing, I decided to check how well it will perform on building new things and I decided to give same exact prompt to both Fable 5.1 and GPT 6 Astra.

Claude Fable 5.1 vs GPT 6 Astra

The prompt, in short, was to create a 3D multiplayer space game that will use real world physics rules to further advance a civilization and will use AI which will learn from the players and be the players when the players are offline.

GPT 6 Astra worked for 20 minutes and Claude worked for 28 minutes after which it finished a Phase 0 as it called 0 where I had only login. Had to add another prompt to tell it to continue and after that it worked for 1.5 hours. Below is a video of what was achieved by both (left is Fable 5.1 and right is GPT 6 Astra):

After this I tried one more thing that I’ve been seeing people building online and that is a Human Atalas.

I sent the following prompt to GPT 6 Astra Ultra and as you can see there is no structure or anything to the prompt, just what it came to my mind at that moment:

Create human atlas for male body. Expand collapse all organs and option to choose muscles, organs, full body, nervous system, blood system etc.
Functionality to expand collapse with option to click on things and see what is what. But make this realistic not some cartoon.
Load some skills for that, design, biological research and anything to make this look as real as possible.

And below is the result:

I didn’t have to write anything additional and this took about 40 minutes to make. It is also accessible here: https://human-anatomy-atlas.mja12300.chatgpt.site/

Overall GPT 6 Astra is good and t makes work easier. I like that there is no 5 hour limit. I will continue using it for a month and see how it goes and if it will convince me to switch from Claude.

Conclusion

• GPT 6 Astra is faster
• Has no 5 hour limit
• Well defined prompt will probably do better
• Claude asks more questions if something is not clear, GPT didn’t ask me for any questions
• Computer use is definitely an advantage although I am not yet convinced it can do things as easy as in the demo. I will continue testing this.