Claude Fable 5.2 vs Opus 5.5: which should you use?

Anthropic's launch notes say that for most coding and knowledge tasks, users should start with Claude Opus 5.5, its strongest daily driver, and then switch as required. The switch can go upward to Fable or Mythos, or downward to Claude Sonnet 5.5 or Claude Haiku 5.5.

It also says Fable and Mythos are best for the most ambitious, open-ended technical work. So the default is not Fable. You begin on Opus 5.5 and move up when the job is bigger or vaguer than Opus handles well.

The widely shared tip says Fable 5.2 for planning and Opus 5.5 for execution. It is consistent with Anthropic's wording, but it is a tip and not an Anthropic rule.

On Humanity's Last Exam, a test of expert-level academic questions, Anthropic reports 74.2% for Fable 5.2 with tools and 67.7% for Opus 5.5. In a test where about 1,000 agents formalized the proof of Fermat's Last Theorem in a math programming language, Opus 5.5 completed 26% in 18 hours, and Fable 5.2 completed the full formalization in 30 hours.

Anthropic says Fable 5.2 is a step above Opus 5.5 in almost all the capabilities it evaluated. These are Anthropic's own results on hard technical tests, so they show real headroom for difficult problems. They do not tell you whether your weekly status report needs it.

Early testers said Fable 5.2 shows better judgment on ambiguous, data-heavy tasks. That is what a plan needs: deciding what the real question is, what to ask first, and what order to work in.

Anthropic also says Fable 5.2 can act as an orchestrator, a model that hands simpler tasks to smaller models, and as an advisor that smaller models consult on hard problems. Those are the two directions of this workflow, so Anthropic has at least described the pattern.

What it has not published is a head-to-head test of Fable planning and Opus executing against either model doing everything alone.

In one test run by a trading firm, Fable 5.2 played every needed role in a research setup: code-writer, coordinator, planner, and job-submitter. Earlier models had needed people to rebalance those roles by hand. For some jobs, one strong model doing all of it may be simpler than a handoff.

Cost can also point the other way. Anthropic reports that on one coding test, Fable 5.2 at High effort scored about the same as Opus 5.5's best score for roughly half the cost per task. Compare the cost on your own work before assuming the cheaper-looking model is cheaper.

A handoff also adds work for you. Each step must be written clearly enough for the second model to follow without guessing.

A model reviewing its own plan can repeat its own mistakes with confidence. Fable reviewing Opus's work is better than no review, but both are language models, and they can share blind spots.

Use a source that does not come from either model. That could be the original contract, the real spreadsheet, a company policy, or a colleague who knows the account. Ask the model to point to where in your source each claim comes from, then open the source and look.

This step is free and it catches the errors that matter most, such as a wrong number or an invented quote.

Pick one real task you do regularly, such as a project plan, a vendor review, or a monthly report. Run it three ways: Opus 5.5 alone, Fable 5.2 alone, and Fable 5.2 planning with Opus 5.5 executing.

Score each result the same way. Note how many corrections you had to make, how long it took, and how much of your plan's usage it used. Keep the setup that gives you the best result for the least effort.

One task is a small sample. Repeat it on two or three more before you change how your team works.

Nightschool AI is an independent learning platform and is not affiliated with or endorsed by Anthropic. Claude, Fable, Mythos, Opus, and Sonnet are names of Anthropic's products.