Claude Fable 5.2 or Opus 5.5: Which to Use for What

The short answer

Start with Opus 5.5. Bring in Fable 5.2 for the parts of a job where a wrong plan is expensive. That is not our invention: in the Fable 5.2 announcement on November 9, Anthropic itself recommends starting most coding and knowledge tasks with Opus 5.5, its everyday model, and using Fable 5.2 as an advisor or orchestrator above it.

This is a proposed way of working, not a hands-on review. We have not run these models against the tasks below. The benchmark results Anthropic published at launch are its own runs.

What changed with Fable 5.2

Anthropic says Fable 5.2 improves on Fable 5.1 in coding, reasoning and vision, produces output up to 30% faster, and uses fewer tokens per task at the Medium effort setting. It also refuses far fewer legitimate requests that the safeguards used to block. The price went down: $8 per million input tokens and $40 per million output, against $10 input for Fable 5.1. A fast mode costs twice as much.

Opus 5.5 costs half as much per token as Fable 5.2, on both input and output. That gap is the whole reason to split the work rather than run everything on the bigger model. Our listings for Claude Fable 5.2 and Claude Opus 5.5 link to the sources.

Give Fable 5.2 the decisions

The jobs worth paying more for are the ones where everything downstream depends on getting one step right:

• Breaking a large change into steps, and deciding the order
• Choosing between two designs before any code is written
• Reviewing a finished diff for the mistake nobody asked about
• Untangling a bug that has already defeated a cheaper attempt

These are short in tokens and long in consequence. A plan is a few thousand tokens; a wrong plan costs a day.

Give Opus 5.5 the volume

Once the plan exists, the steps themselves are usually well defined: write this function, update these tests, change this copy, migrate this file. That is where most tokens go, and where paying half per token adds up. If a step keeps failing, that is the signal to send it back up to Fable 5.2, not to switch everything over.

Test the split on your own work

Pick one real task you have done recently, so you know what a good result looks like. Run it three ways from the same starting files and the same brief:

1. Opus 5.5 alone
2. Fable 5.2 alone
3. Fable 5.2 writes the plan, Opus 5.5 carries it out

For each, record whether the result was acceptable, how much you had to fix by hand, how many follow-up requests it took and what it cost. Keep the failed attempts in the record. The cheapest setup that reaches an acceptable result is the one to keep, and it will not be the same for every kind of task.

What to watch for

On Claude plans, Pro and standard Team seats can spend at most half of their weekly usage on Fable, so a plan-and-execute split also stretches your limit. Fast mode doubles the price; use it where waiting is the real cost, not by default. And Claude Mythos 5.2, released the same day, is the same model with fewer cybersecurity and biology safeguards. It is only for vetted organisations, so for most people the choice really is Fable 5.2 or Opus 5.5.

Read the November 9 briefing for the other releases that day, including Meta's low-cost Muse Ember, built for exactly this kind of worker role.