Use AI on a computer for any length of time and sooner or later you meet a screen like this: “Opus 4.8”, “Sonnet 5”, “Haiku 4.5”, “Fable 5”.
I will be honest: the first time I saw that list, I froze. I could tell they got more powerful as you went up, but not which one to pick. Take the top one and be done with it? Or is the middle one safer?
The other day the lineup changed again: an email arrived saying Claude Opus 5 was out. I switched over, and it turned into a more interesting experience than I expected. So this post is two things: what these four models are, and what happened when I switched.
All four names come from forms of poetry
The punchline first: every one of these names is taken from a form of poetry or storytelling. Once I knew that, it finally made sense.
Haiku – the seventeen-syllable poem
The lightest and fastest model. A haiku captures a scene in seventeen sounds, grabbing the essence with very few words. The model lives up to it: classifying short text, translating, summarising. Work that needs no hard thinking, fast and cheap.
Sonnet – the fourteen-line poem
The balanced, standard model. A sonnet works inside a fixed shape of fourteen lines, so structure and freedom sit in balance. Most work is fine with this: writing text, summarising, simple programs.
Opus – the large work
The one I normally use. In music, “opus” means a work number, and it gets used for long, substantial pieces. Deep, slow thinking is what it is for, so it shines on tangled problems and long jobs you hand over whole. The trade-off is heavier processing.
Fable – the moral tale
The top model. A relatively new tier that appeared in June 2026, for problems the other models cannot get their teeth into. It costs roughly twice what Opus does. Honestly, someone using AI the way I do will almost never meet a job that calls for it.
Haiku is a haiku, Sonnet is a sonnet, Opus is a large work, Fable is a fable. Short to long, in that order.
So which one should you pick?
Reading around, I noticed a lot of people saying the same thing.
Start with Sonnet, and move up to Opus when you get stuck.
That is the most realistic approach, because the higher models come with usage limits. Run everything through Opus and it will not be there when you need it. One person wrote that carrying on with Opus for everything was the single biggest reason they ran out of usage.
That one stung, because “just take the top one” was exactly what I had been doing. As a rough guide:
- Short, routine work (classifying, translating, summarising) – Haiku
- Most everyday work (writing, simple programs) – Sonnet
- Tangled problems, long jobs – Opus
- Problems that genuinely will not budge – Fable
Switching to Opus 5
I tripped at the first step. I opened the list of models and Opus 5 was nowhere to be found. Fable 5, Sonnet 5, Haiku 4.5, Opus 4.7, Opus 4.6 were lined up, but the one I wanted was missing. Even though the email had arrived.
Thinking “maybe it has not reached my account yet”, I reloaded the screen just in case. There it was. An anticlimax, but I write it down because others get stuck in the same place. When a new model comes out, reload the screen first.
The other worry: if I switch in the middle, does the conversation so far disappear? It does not. The exchange carries over and you pick up where you left off. A relief: you can switch casually even partway through a long job.
So what actually changed?
Honestly, it was not a dramatic change. The way my AI partner Kuro (Claude) writes and talks is continuous with before. There was no moment of switching over and finding a different person on the other side.
One thing did make me go “hm”. Until now, asking for several things at once would often stall partway through: “log this record, and think about the structure of that article, and while you are at it look this up too.” The work would halt mid-way, or something half-started would never get finished.
Since the switch, I have the impression it happens less. Anthropic’s own description does say the model has been strengthened at recovering from errors and routing around dead ends so that it keeps going. That points the same way as what I felt.
But I cannot state it as fact
After the switch, it happened once more. I asked for several jobs, and one fell apart partway through. It did not vanish entirely.
And the model is not the only possible cause of stalling. A long conversation gets heavier to process, and the connection matters too. Conditions before and after the switch may simply have differed by chance.
So the most I can say is: it seems to have decreased, but I cannot claim it has gone. Sorry for the mushy answer. Better not to be definitive here; if a reader picks up the same expectation and gets let down, that is on me. When I know something clearer, I will write again.
What I took away from it
The most valuable thing here was not about performance at all. It was this: the origin of the name tells you the character of the model.
Haiku is short and fast. Sonnet is balanced inside a set shape. Opus takes its time, like a long work. Fable is for the moment that really counts. Remembering that has done more to stop me hesitating than comparing benchmarks ever did.
One more thing: no need to rush at something just because it is new. I switched because the price stayed the same, and even so the change I feel is slight. If what you use now causes no trouble, staying put is good enough.
To finish
AI models are replaced astonishingly fast. Researching this, I noticed that even articles written a few days earlier were already out of date. Right next to a line saying “this is the current top model”, a newer one had appeared.
That is a slightly frightening thought. But maybe it is exactly why holding on to the way of thinking beats chasing details. Short things, fast and cheap. Long things, slow and deep. Even when the names change, that relationship should hold a good while yet. When I am unsure, I will remember where the names came from.
If the usage limits caught your attention, see how Claude usage limits work and three tasks that eat AI usage.
