Opus, Fable, Sol, Astra... which AI model do I actually pick?
Chris Rupp
Your agent has been stuck on the same bug for twenty minutes. A friend texts you “switch to Fable.” So you open the model menu and there’s Haiku, Sonnet, Opus, Fable… or if you’re in Codex it’s Luna, Sol, Astra. Which one are you supposed to pick? And is your friend right, or did they just pick the biggest name on the list?
They’re just sizes, from two companies
This is the part that clears up most of the confusion. These aren’t a bunch of totally different products, they’re sizes. Anthropic makes Claude, and OpenAI makes GPT, and each company sells its model in a few sizes.
Put them all on one ladder and it goes like this… Haiku and Luna at the bottom, then Sonnet, Sol, Opus and Astra, with Fable at the top.
The lower you go the faster and cheaper it is. The higher you go the deeper it thinks, but it’s slower and it costs more every time you use it.
The names actually hint at the size
Anthropic named theirs after writing. A haiku is three lines, a sonnet is fourteen, and an opus is a whole symphony… so the longer the piece, the bigger the model. Then Fable comes along and breaks the pattern, cuz a fable is a short story but it’s the biggest Claude there is. Just one of those things you have to memorize I guess.
OpenAI’s names go out into space instead. Luna is the moon, Sol is the sun, Astra is the stars. Small, bigger, biggest.
Why picking the wrong size costs you
You’d think you should just always pick the biggest one, right? Not really.
Put Fable on a typo and you wait longer and pay more for the exact same one word fix a small model would have done in seconds. Go the other way and put Luna on a tricky bug, and it can get lost and go around in circles, which honestly might be why your agent has been stuck for twenty minutes in the first place.
So the move is to pick by how hard the job is:
- Most coding: start in the middle, on Sonnet or Sol. They handle the bulk of everyday work just fine.
- Tiny stuff: renaming a button or fixing a typo? Haiku or Luna is plenty.
- Stuck on a hard bug: step up to Opus.
- The really hard jobs: that’s what Astra and Fable are for, so save them for when you actually need them.
So yeah, your friend wasn’t wrong exactly… but you probably wanna try Opus first before you jump all the way to the top.
One catch… the ladder keeps moving
These companies are racing each other to put out new models, so the names in this article could be old news in a month. Before you pick, take a second to check what’s current. The idea stays the same though, small and fast at the bottom, big and slow and pricey at the top.
And one more thing that trips a lot of people up… Claude Code and Codex are not sizes. They’re the agents, the apps that actually go in and edit your files. The model you pick is the brain doing the thinking inside them. To switch models in either one, just type /model.
How I stopped picking a model every single time
This is the part that got old for me fast, switching models by hand every time the job changed. So in Orbit, the program I built to run my agents, there’s Model Tiers. You set each tier to a model once… Low for quick edits, Medium for everyday features, High for hard changes, and Ludicrous for the biggest jobs. Claude or GPT, that’s your call, and you can even point a tier at a local model if you run one.
Every prompt in your queue gets a tier, so it goes to the right model on its own. Typo fix goes to the cheap fast one, the gnarly refactor goes to the big one, and I’m not sitting there flipping menus. And when a new model comes out, I just swap it into its tier once and everything after that uses it. Orbit is in Alpha and free right now if you wanna try it: https://orbitdev.app
All of these words, plus the ones that come next on the ladder, are in the Learning Atlas if you want to go deeper.
What’s your default model right now, and have you ever caught yourself using the big one on something tiny? I’m guilty of it more than I’d like to admit.