Sonnet 5.5 vs Opus 5.5 vs Fable 5.1: Which Claude Model?

Which Claude model should you use: Sonnet 5.5, Opus 5.5 or Fable 5.1?

Use Sonnet 5.5 for most everyday work, especially tasks with clear instructions. Switch to Opus 5.5 when the goal is vague or the job is long and messy. Reserve Fable 5.1 for demanding reasoning and long autonomous runs, since it costs 2.5x what Opus does on pay-per-use pricing. Haiku is the cheapest and fastest option for small, quick jobs. Anthropic's own release notes say Opus 5.5 performs near Fable level on most work.

What are the Claude model tiers?

Claude is a family of models, not one AI. Right now the main three share the 5.5 version number, with Fable sitting above them.

A useful way to think about it is staffing. You would not have your most senior person answer every quick email, and you would not hand a whole project to your newest hire. Match the model to the weight of the job.

How did Sonnet 5.5 compare with Opus 5.5 in a real test?

One creator ran Sonnet against Opus on seven real workflows. The pattern was consistent enough to be a decent rule of thumb, though it is a single person's test, not a formal benchmark.

• Clear instructions: Sonnet won. A resource guide, for example, finished in about half the time and usually cost less than the Opus run.
• Vague goals: Opus won. A 90-day plan, where the model had to decide what "good" looked like, came out better from Opus.

The logic is straightforward. When you spell out the steps, a bigger model has little extra to add, and you pay for speed and tokens you do not need. When the brief leaves gaps, the stronger model fills them with better judgment.

For most document and general knowledge work, the same creator suggests Sonnet 5.5 on medium effort, particularly when combined with skills.

Does effort level matter as much as model choice?

Effort controls how hard Claude thinks before answering. The levels are low, medium, high, extra high and max. On the 5.5 models the default is medium. You change it with /effort or through the effort selector at the bottom right of the desktop app.

More is not always better. In one experiment, Opus 5.5 got the identical prompt at each level:

Anthropic's own docs reportedly warn that max is prone to overthinking. The practical takeaway is to keep medium as a daily default and raise effort only for hard work. The creator uses extra high for video editing, for instance.

There is also Ultra Code, which is less a level and more a setting where Claude plans the whole workflow before starting a big task.

What does Fable 5.1 cost, and is it worth it?

Fable 5.1 is Anthropic's top-tier model, built for demanding reasoning and long tasks where Claude works on its own. It is also expensive. On pay-per-use, it runs 2.5x the price of Opus.

Plan access works like this:

• Max and Ultra: Fable can consume up to half of your weekly limit.
• Pro: You pay for Fable with usage credits.

Whether it is worth it depends on the job. Given Anthropic's statement that Opus 5.5 performs near Fable level on most work, Opus is plenty for most use cases. Fable makes sense when a task is hard enough that Opus visibly struggles, or when you want a long, unattended run to be as reliable as possible. For anything routine, you are paying a premium for headroom you will not use.

How does the /advisor command work?

The /advisor command lets Claude check in with a stronger model, such as Fable, at key moments during a task. The main work runs on a cheaper model. The expensive one weighs in only where judgment matters most.

That gives you a middle path. You get some of Fable's reasoning without running the entire job on it and without burning through half your weekly limit. The sources describe the command only at this level, so test it on a real task to see how often it escalates and what that does to your usage.

How do tokens and plan limits affect model choice?

Tokens drive both cost and how fast you hit limits. Every current model in Claude Code has a 1 million token context window, which Anthropic says is roughly 555,000 words.

Most tokens go to reading, not writing. Across seven tasks the creator measured, Claude read about 160 million tokens and wrote about 800,000. That is roughly 200 read for every one written. So the files and instructions you load matter more than prompt length, and a heavier model reading all that context costs more.

Limits run on two clocks at once: a 5-hour session limit and a weekly limit. Chat and Claude Code share the same bucket. Plans as described:

• Free ($0): Sonnet and Haiku, no Opus.
• Pro ($20/month): Claude Code included.
• Max ($100/month): five times Pro usage.
• Ultra ($200/month): ten times Pro weekly usage, and verified accounts no longer have a 5-hour limit.

Type /usage to see where you stand.

A simple way to pick a model

1. Start with Sonnet 5.5 on medium effort for anything with clear instructions.
2. Move to Opus 5.5 when the brief is fuzzy, the task is long, or Sonnet's output misses the point.
3. Raise effort before raising the model, but stop at extra high unless you have evidence max helps.
4. Try /advisor when you want a stronger second opinion at critical steps.
5. Choose Fable 5.1 only for the hardest, longest autonomous work.