Use Opus 5.5 for most daily work, and consider Fable 5.2 when Opus repeatedly falls short. For a reproducible bug, a feature with clear requirements, or a report based on supplied documents, start with Opus. If it keeps failing the same tests or overlooking key constraints despite having the necessary context, try the same task with Fable.

Before switching models, try increasing Opus’s effort level. Anthropic recommends starting most work with Opus 5.5 at medium, then considering Fable if Opus at high remains insufficient. For particularly demanding work, such as a major migration or a difficult architecture decision, starting with Fable is also reasonable. Anthropic’s model selection guide

Fable costs twice as much per token, so judge it by the work it saves. Its standard API input and output rates are $8 and $40 per million tokens, versus $4 and $20 for Opus. The upgrade is worth considering when it reduces failed attempts and manual corrections enough to offset the higher rate. Official pricing

Based on official documentation and published benchmarks; no independent hands-on testing.


Opus 5.5 vs Fable 5.2 at a glance

Specification

Claude Opus 5.5

Claude Fable 5.2

Standard API input price, per million tokens

$4

$8

Standard API output price, per million tokens

$20

$40

Cache read price, per million tokens

$0.20

$0.20

Context window

1 million tokens

1 million tokens

Default API effort

Medium

High

These are API rates, rather than subscription prices. Cache writes and other billing options have separate rates. Opus 5.5 specifications, Fable 5.2 specifications

Both models accept the same amount of context. Moving to Fable therefore does not give you a larger document allowance; the question is whether it makes better use of the information you provide.

For a practical starting point:

Your situation

Suggested approach

Opus already produces work that passes review

Keep using Opus

The answer misses requirements or skips verification

Fix the instructions and context, then test higher effort

Opus repeatedly fails despite adequate information

Compare against Fable

The project involves difficult architecture decisions or interacting constraints

Include Fable in the initial evaluation

You repeat a task hundreds of times

Compare total cost per accepted result

Only one stage requires difficult reasoning

Test a workflow that uses both models

When to choose Claude Opus 5.5

Opus is a sensible starting point when you can describe the outcome clearly and judge whether the result meets it.

Examples include implementing a feature against an existing specification, investigating a reproducible bug, reviewing a proposed change, or preparing a report from supplied material.

This does not mean Opus is limited to small jobs. Anthropic positions it for multihour coding agents, large refactors, and complex systems engineering. Project size alone is a poor reason to switch. Opus 5.5 overview

Consider a checkout change with established business rules and existing tests. Before trying a more expensive model, give Opus the relevant code, the expected behavior, and a clear definition of completion.

Then examine the result:

  • Does the implementation satisfy the requirements?

  • Does it preserve behavior outside the requested change?

  • Do the relevant tests pass?

  • How much correction does a reviewer need to make?

If the answers are consistently satisfactory, you have little reason to change models for that workflow.

Sources & references