Hi friends,
We are back after a short two week summer break. Lots happened over the last 2 weeks. Fable 5 is back, though only temporarily till July 12. Sonnet 5 shipped. Opus 4.8 is still around. And GPT-5.6 launched yesterday as well.
They all overlap. So which one to actually use? That's what I've been digging into: which model to use, and when.
Which model for which job
The tempting answer is to use the best one for everything - point Fable at every task and stop thinking. I think that mostly works (till July 12) as long as you use Claude Code on a subscription.
However, it is also the most expensive way to use Claude Code. Fable is $10/$50 per million tokens. Opus is $5/$25. Sonnet 5 is on a promo $2/$10.
The people who've sorted this out don't pick one model. They give each one the job it's good at. Fable plans. A cheaper model builds.
Here's what I use each for:
Fable for the vague, long, multi-file stuff. Its lead grows with difficulty. It's also slow and twice the price, so don't waste it on small work.
Opus 4.8 if Fable is not available (or I am running low on usage limits).
Sonnet 5 as the cheap executor.
GPT-5.5 if you want the strongest builder and don't mind a second subscription. It's the one a lot of people reach for to grind out implementation once Fable has written the plan.
I have no way to prove that is the optimal setup. There's no clean benchmark that says "use Fable for X."
It's mostly using the models for different tasks, looking at the output, forming a gut read. Vibes, mostly. I’d be curious to hear which models y’all are using for your day to day work.
A few other things worth knowing
Sonnet 5's promo pricing runs through August 31: $2/$10 per million tokens, with a native 1M-token context window.
The memes
That's the week. Let me know what you guys are building with Fable before it goes away!
We will be back next week with some more deep dives.
Thanks!
Abhishek