I migrated to Lviv for now because russian fascists were bombing my lovely home city Kharkiv. Just if you wish to know. I'm safe, working and very angry.
Remember using the t3 code 3 months ago, and it was a little behind the Codex app (or ChatGPT now), so I switched back. Now the t3 code feels like a mile ahead. I want to thank a lot to @jullerino for the amazing work you've done.
My theory is that SOl 6 is actually a smaller model and maybe renamed Terra 6. I see @OpenAI doing it to make it scalable for the ultrafast tier. It would be amazing to use it with, like, high or xhigh if quota allows and it's efficient. I will not pay 500 for it, maybe 200$
@maurice_build Sol 6 is like 5.6 but a little bit different; I would say in a better way. I wouldn't call it dumper outright. I use recommended Astra agents.md which makes it behave more like 5.6, so consider it. I recommend Sol 6 for easy tasks, but it's nowhere near Astra. Use Astra low
@melvindvivas Getting all the good of Astra without being meaningfully better, but cheaper. Good enough, But it seems I will be Sol Xhigh-Max user now, which is very slow
I switched from Windows to Bazzite for my gaming/programming-server PC, and that's pretty solid, works well, has some caveats. But how the F is it bad? It feels like iOS. What do you mean I can't just download the package, install it and use??? I need to create bottles???
@thsottiaux Last time I used my Pro limits, I got Luna Reserve. Now I didn't get Luna Reserve and can't even subscribe to 20x instead of 5x. Adorable experience...
I used Pro x5 sub on ChatGPT for 3 months, and I just thought about the idea of getting x20, but now it's too late. Hopefully we will get at least some new 6 model (Sol or any), so the Astra demand will be lower. Anyway, you cannot use just Astra 5 workdays long even on x20
@thsottiaux
I have access to Astra inside my ChatGPT app and Codex CLI on Linux, but don't have it with the same account on my ChatGPT app and CLI on my Mac...
Everywhere is the latest version; I've checked
The only weak part of Sol-5.6 in the ChatGPT app today is the default context limit, because with a 256k context limit today it's definitely less efficient and often compacts too much.
It was definitely built around a 350k context, and I hope the @OpenAI team will fix the issue
@thsottiaux When remote connected from mac to windows pc using 2 chat gpt work apps, you cannot open almost any file from windows on mac, even download it.