Two big AI models came out in the same week this month, Claude Fable 5.1 from Anthropic and GPT-6 Astra from OpenAI. I've used both every day since.
People keep asking me which one is better. I use both, but for different jobs. Fable writes my code. Astra runs my computer.
Why I code with Fable
Fable has better taste. Way better.
Astra is a good model and its code runs, but when I read it, something is missing. It's hard to point at. The code works and it still doesn't feel like it belongs in my project.
Here's what I think is going on. Every project has rules, like how we name things and which parts to reuse. I also write skills and rule files so my coding agents know how I like things done. Fable reads all of that and follows it, but Astra often skips it. It does the task I typed and forgets the rest, so I end up saying the same thing twice.
I'd even put Opus 5 ahead of Astra here. Opus 5 is Claude's smaller, cheaper model, and it still gets what I'm asking for more often.
Astra does beat Fable on some public coding tests, and I believe those scores. But a test only checks if the task got done. It can't tell you if you'd want that code in your project.
This matters a lot for me because I run a few coding agents at the same time on client builds. If one of them ignores my rules, the mess grows fast. With Fable I clean up less.
What they cost
On the price list they cost the same. Both charge $10 for every million tokens you send in, and $50 for every million that come out. A token is a small chunk of text, a bit shorter than a word.
The difference shows up in two places.
The first is re-reading. A coding agent reads the same files again and again while it works. Both companies give a discount on text the model has already seen. With that discount, Fable charges 25 cents per million tokens and Astra charges a dollar, which is four times more.
The second is long sessions. Send Astra more than 272,000 tokens at once and its input price doubles. Fable's price stays the same however long the session gets.
My coding sessions are long, so Fable comes out cheaper for me.
One fair warning. Astra needs fewer tokens to finish a task, so on small quick jobs it can be the cheaper one.
There's also a cost no price list shows. When I have to rewrite a model's code, that's my time. I rewrite less with Fable.
How I use them together
New problem? Fable builds the first version. That's where the big choices get made, like how the parts fit together and what happens when something breaks. I want the best model for that.
After that I switch to Opus 5 for the small stuff, like fixes and tweaks. Opus 5 costs half as much as Fable ($5 in, $25 out), and it's plenty good for that kind of work.
Where Astra wins
Astra is better at using a computer. I mean clicking through apps and browser tabs the way a person would. GPT-5.6 Sol, the OpenAI model before it, was already good at this, and Astra is better again.
I use it a lot for work that isn't code. It finds leads and helps with my email outreach. It also does the boring clicking in tools that have no API.
Fable can use a computer too. Astra is just better at it, so that work goes to Astra.
So which one?
If you write code, go with Fable 5.1. Use Opus 5 for the small fixes and you'll save money. If you want an AI to do the clicking for you, go with Astra.
I use both, and I don't see that changing soon. New models ship every few months, and I try each one the day it comes out. When my setup changes, I'll write about it.