I build software that gun stores run on | 馃嚭馃嚫 | Tweeting about development, marketing, and travel. 馃嚮馃嚘
Tampa, FL
Joined November 2015
- Tweets3.9K
- Following3.4K
- Followers8K
- Likes113K
Obviously @thsottiaux isn't going to be giving out resets like before - but it'd be nice if they at least opened up the ability to buy resets like they trialed for a subset of users about a month ago. Because, this is not ideal...
We've been pushing some serious volume through GPT-6 Astra. A few things I've noticed when comparing to Fable.
1. Most of what I hear about usage is noise. It's obviously more expensive than Sol, but you still get great value when compared to Fable 5.1 and their 50% limits. If Anthropic removed the 50% limit, it'd be a lot closer.
2. It stops, a lot. I didn't notice this as much with GPT 5.6 Sol, but I often run into issues where the model will stop after a planning phase and ask me whether it should proceed before it starts execution on a simple task that I told it to just go and do.
3. I don't have any real data to back this up, but the way that it goes and uses the plugins that it has access to (Gmail and Wispr Flow primarily for me) seems significantly more efficient than 5.6 Sol.
4. Seems to be a common critique, but the code it writes is still less readable than Fable 5.1, but it does seem more robust in a few ways that matter for my use cases.
5. Its ability to ask questions in the middle of its response and then keep working on things that don't require answers to those questions is very, very cool.
6. Frontend design is still not as good as Anthropic, but they're closing the gap some.
7. They seem to have done a lot of work to make the responses that it gives clearer. I don't notice nearly as much elaborate prose in the answers it gives me when I ask it a question.
