Use Claude Fable 5 for any task that is both difficult and important. Use Opus 4.8 for hard work that doesn’t clear that bar. Use Sonnet 5 for simple tasks like drafting an email.
At TJ Digital, we run about 90% of our daily work through these models across roughly 42 active client campaigns. My team has tried every model and effort level across hundreds of tasks over the past few weeks.
The bigger mistake I see is people picking the right model and then using it the wrong way. Fable rewards a completely different prompting style than Opus does. Most people never change how they prompt, then conclude the expensive model wasn’t worth the money.
Which Claude Model Should You Use for Each Task?
Here’s the routing table we use internally.
| Model | Use it for | Effort level | API price per million tokens (input / output) |
| Fable 5 | Tasks that are both difficult and important | High only | $10 / $50 |
| Opus 4.8 | Most difficult work, and anything Fable is overkill for | High by default, Extra or Max when it matters | $5 / $25 |
| Sonnet 5 | Simple, well-defined tasks like drafting an email | Low or Medium | $3 / $15 |
| ChatGPT in Codex | Web research, image generation, computer use | Default | $5 / $30 |
I’m leaving Gemini off that table. I don’t think you should be using it for work right now.
When Should You Use Claude Fable 5?
Use Fable 5 when the task is difficult and important. Difficult but unimportant work wastes the token price. Important but easy work doesn’t need Fable at all.
Some people will tell you Fable is no better than Opus. They’re wrong. My best guess is that they aren’t respecting how smart Fable is, so they’re still treating it like a dumber model.
The benchmark gap is not subtle. On SWE-Bench Pro, Fable 5 scored 80.3% against GPT-5.5 at 58.6%. Anthropic built this model for work that takes a person hours, days, or weeks to finish.
How Should You Prompt Fable 5?
Tell Fable the goal and let it work out the steps.
With older models, you had to tell them what to do and exactly how to do it. You gave rigid step-by-step instructions and a lot of reminders, or they went off the rails. Since around Opus 4.6, we’ve been able to tell the model what we want and let it figure out the sequence.
Fable moves the line one step further. With Fable, you state the outcome you want and stop there.
Say you want a dashboard. Tell Fable what the dashboard will be used for and what it needs to accomplish. It will make better decisions about what goes on it than you would have.
That shift is the whole reason people come away disappointed with Fable. They hand it a step-by-step script written for a dumber model, then wonder why it performs like one. I covered this in more depth in how AI prompting changed over the past six months.
What Effort Level Should You Use with Fable 5?
Use High. That’s the whole answer.
Go lower and you get a dumbed-down version of Fable. Go higher and you’re burning tokens for nothing. Fable Extra, Max, and Ultra Code all run the same model, they just spend more tokens on the problem.
Ultra Code runs more checks and spins up more adversarial agents. About 95% of the time you get the exact same result you’d have gotten on High. Unless you work at Anthropic and have unlimited tokens, that trade isn’t worth making.
When Should You Use Claude Opus 4.8?
Opus 4.8 is the best default for most work. It costs half of what Fable costs. It’s also what you fall back to when Fable isn’t available to you, which is most of the time for most people.
Effort levels work differently here. With Opus, it’s worth moving up to Extra, Max, or even Ultra Code on hard, important tasks. We see a real jump in performance when we do.
My read on why is simple. Opus makes more mistakes than Fable does, so making it double-check its own work actually pays off. Just make sure the task is worth the tokens you’re about to burn.
| Task | Fable 5 | Opus 4.8 |
| Hard and important | High effort | Extra, Max, or Ultra Code |
| Hard but routine | Skip it, use Opus | High effort |
| Effort above High | Wasted tokens | Real quality gain |
For everything else, Opus on High is a great default. I wrote a full breakdown of Opus 4.8 when it shipped.
When Should You Use Claude Sonnet 5?
Use Sonnet 5 for easy tasks. If the work is truly simple, like drafting an email or cleaning up a list, Sonnet is the cheapest way to get it done.
Here’s the thing about Sonnet. Hand it something difficult and it burns so many tokens working the problem that it ends up costing you about what Opus would have cost anyway. You paid the cheap rate and got the cheap result.
Anthropic’s own numbers show Sonnet 5 landing close to Opus 4.8 on many tasks, at 80.4% against 82.7% on a coding benchmark. That closeness is exactly what fools people.
The gap opens up on the hard end. Crank Sonnet to its highest effort level to close that gap and you stop saving any money. Our Sonnet 5 review goes deeper on where the line sits.
Where Does ChatGPT Fit in a Claude Workflow?
ChatGPT inside Codex is the best model for a few specific jobs, and on a Codex subscription it’s incredibly cheap.
The move I recommend is having Claude call the Codex CLI on your behalf. Tell Claude to hand off three kinds of work:
- Computer use
- Image generation
- Deep research across the internet
Claude stays the brain of the operation. Codex does the legwork it’s better at.
ChatGPT is also a good fallback when you’re running out of tokens and the task isn’t especially difficult or important. Most teams hit their weekly limit before the week is over.
How Long Will Fable 5 Stay on Subscription Plans?
Fable 5 is included on paid Claude plans through July 19, 2026, for up to 50% of your weekly usage limit. After that, using it requires prepaid usage credits at $10 per million input tokens and $50 per million output tokens.
That date has now moved three times. Anthropic extended access through July 19 on July 12, hours before the previous cutoff. When I recorded my video on this, I said the deadline would probably move again, and it did.
Plan around the date you actually have. Bank the difficult, important work on Fable while it’s included in your plan. Route everything else to Opus and Sonnet.
What Else Should You Know Before Switching Models?
Is Claude Fable 5 actually better than Opus 4.8?
Yes, on hard problems. Fable holds a longer context, checks its own work more thoroughly, and dispatches parallel subagents more reliably. On easy problems you won’t see the difference, and you’ll still pay double.
Is Opus 4.8 good enough if I never touch Fable 5?
Yes, for almost everything. Opus 4.8 on High handles most difficult work, and moving it up to Extra or Max covers the hardest tasks. Fable earns its price on long, open-ended projects where you hand over a goal and walk away.
What effort level should I use with Claude Sonnet 5?
Low or Medium. Sonnet defaults to high effort, and the entire point of Sonnet is cost efficiency. Stepping the effort down keeps that advantage intact.
Is Claude Ultra Code a separate effort level?
No. Ultra Code pairs the Extra effort level with permission to spawn multi-agent subtasks. The underlying model is the same one you get on Extra.
How Should Your Business Handle AI Model Selection?
Most businesses run everything through one default model with one default prompting style. We pick the model per task across 42 client campaigns. That discipline is part of how we deliver about four times the work at the same rates as a traditional agency.
Contact TJ Digital and we’ll show you what that looks like for your business.