OpenAI just released Astra, which they call their most powerful model yet.
What caught my attention is not only the coding performance, but the focus on browser/computer use and cybersecurity. OpenAI claims Astra is their best software engineering model so far, with stronger results for bug finding, terminal tasks and understanding large codebases.
There is also some controversy around Astra using more opaque reasoning techniques, which could make model behavior harder to monitor as capabilities increase.
It will reach ChatGPT Plus, Pro, Business and Enterprise users during the next week, together with API access.
What are your results with GPT-6 so far?
Iām just getting started with it and quite early to form an opinion, but noticing so far:
- Seems to eat a lot of tokens
- Iām not convinced itās better than 5.6 for my goals - I feel like itās making too many changes / too aggressive, whereas I like how 5.6 is very balanced.
- It does seem to be quite smart, implemented a fairly large task quickly and accurately
We will see, for me GPT 5.6 is more than enough, either way AI is evolving like crazy, remember when we used to code by hand and search on forums to fix a stupid bug, well it was not that long ago believe it or not 
I still use 5.6 mainly, however, my gpt-6 testing was as follows: it seems to be more aggressive than 5.6, it changes things more than requested and seems to consume the token limit much faster. Interestingly, benchmarks from around the net say that it uses fewer tokens per completed coding task, so perhaps it simply does more work per run as before. It likes to use that new āobscureā multi layer reasoning.
For me, for my everyday WordPress work, 5.6 still seems to be much better. Too early for a final verdict, though, as it might also change under the hood without notice, as OpenAI did this in multiple cases so farā¦
You need to test it on a pro plan, the agents are different that the regular plan, not sure how it works but the pro plan 100+ $ bahves much better than the low tear one in all regardsā¦
I tested Astra today, is very good, I mean it feels like it know what I want, it also surprised me to open the browser and acutally testing some viusal thing for me, impressive, this s*it is starting to be scary, I feel like the dizaster is really close if it continue like this and I am not talking abhout jobs!
Codex (5.6 too) can now visually inspect sites. It can do things like run curl commands etc.
So you can tell it to fix some random visual bug or flicker etc or to test a site or anything like that.
Yesterday I Had it fix a program that was throwing an error on my MacBook, I just told it to fix it and it found the registries cache entries whatever was broken and just fixed it in minutes.
The world is changing very fastā¦
Considereng where we were six months ago this is really impressive!
I donāt know if Iām correct about this but it feels like thereās been an acceleration lately in how many people are using AIs.
Last week I had 3 people report the same bug, clearly discovered using AI (some API thing that a real human would have never tested).
Donāt you know your grandma is a coder now? 
She still makes cookies, but now you have to accept them first. And if you misbehave, youāre getting blocked in real life AND in the firewall. 
This is getting so stupidā¦ if we continue like this software quality will fall so bad it will become a standard!