Beta
Download

Claude Fable 5.2 vs Opus 5.5: which to use for coding

AutoRouterUpdated 10 min read

Key facts

As of November 10, 2026, keep everyday coding on Claude Opus 5.5, and move a task to Claude Fable 5.2 when it is a major refactor, a migration, a long unattended run, or work where Opus 5.5 at high effort still falls short. That is Anthropic's own advice, published with Fable 5.2 on November 9, 2026. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, twice Opus 5.5's $4 and $20, and both charge $0.20 per million tokens for cache reads. AutoRouter, the Mac app for Codex and Claude Code, wrote this post and worked the prices through: one assumed agent request costs $0.139 on Fable 5.2 and $0.074 on Opus 5.5.

When is Claude Fable 5.2 worth using over Opus 5.5?

Claude Fable 5.2 is worth its price on the hardest and longest technical work, and Claude Opus 5.5 is the better choice for most daily coding. Anthropic's models overview describes Fable 5.2 as the model for "demanding reasoning and long-horizon agentic work" and says to start most workloads on Opus 5.5. Its guide to choosing a model says to adjust effort before changing models: raise Opus 5.5 from medium, Claude Code's default effort for both models, to high, and once you would go past high, test Fable 5.2 at a lower effort instead.

Kind of workStart withWhy (Anthropic's guide, November 9, 2026)
Everyday engineering, debugging and feature work across several pull requestsOpus 5.5 at medium, then highThe guide starts this work on Opus 5.5, at half Fable 5.2's price per token.
Major refactors, migrations and architecture planningFable 5.2 at medium, then highThe guide lists all three under Fable 5.2, with leading many subagents.
Long unattended runsFable 5.2The guide says to move up for them.
A hard task where Opus 5.5 at high still falls shortFable 5.2, at a lower effort than you used on Opus 5.5The guide warns that an effort level above high can cost as much as the bigger model.
Computer use and work heavy on screenshotsOpus 5.5 at mediumThe guide describes these long agentic loops as work Opus 5.5 is built for.
You aren't sure yetOpus 5.5 at mediumIt is Claude Code's default model on every plan and on the Anthropic API.

Test each row on your own repository. Which model to use for which coding task covers the models below Opus 5.5 and OpenAI's lineup.

How much does Claude Fable 5.2 cost compared with Opus 5.5?

Claude Fable 5.2 costs twice as much as Claude Opus 5.5 for input, cache writes and output, and the same for cache reads, as of November 10, 2026 (pricing). Both models bill their full 1M-token context window at one rate.

List price, November 10, 2026Claude Fable 5.2Claude Opus 5.5
Input, per million tokens$8$4
5-minute cache write, per million tokens$10$5
1-hour cache write, per million tokens$16$8
Cache read, per million tokens$0.20$0.20
Output, per million tokens$40$20
Fast mode, input / output per million tokens$16 / $80$8 / $40
Default effort on the APIhighmedium
Default effort in Claude Codemediummedium

A coding agent sends the whole conversation with each request and reads most of it from the cache, so one request costs a little under twice as much on Fable 5.2. The example prices the assumed request AutoRouter's posts use: 45,000 cached input tokens, 5,000 new input tokens written to the 5-minute cache Claude Code uses with an API key, and 2,000 output tokens, thinking included.

One assumed agent request, list prices as of November 10, 2026

Claude Fable 5.2
  45,000 cached input  x  $0.20 per million  =  $0.009
   5,000 new input     x $10.00 per million  =  $0.050  (5-minute cache write)
   2,000 output        x $40.00 per million  =  $0.080
  Total                                         $0.139

Claude Opus 5.5
  45,000 cached input  x  $0.20 per million  =  $0.009
   5,000 new input     x  $5.00 per million  =  $0.025  (5-minute cache write)
   2,000 output        x $20.00 per million  =  $0.040
  Total                                         $0.074

The Fable 5.2 request costs 1.9 times as much, a difference of $0.065. At this request size, $20 of API usage pays for 143 requests on Fable 5.2 or 270 on Opus 5.5. The cached part costs the same on both models, so a session that mostly re-reads its context narrows the ratio, and a task that writes a lot of code moves it toward 2 times.

What does a finished task cost on each model?

A finished task can cost less on Claude Fable 5.2 when it is hard enough that Claude Opus 5.5 needs its highest effort, and it costs close to twice as much, for the same tokens, when Opus 5.5 would have finished it at medium. Two results published by November 10, 2026 show the first case:

The first result is Anthropic's own, and the Vals Index mixes coding benchmarks with others such as a finance agent test. To price your own work, run one hard task from your backlog on both models and compare what /usage shows.

On everyday tasks, AutoRouter has measured Opus 5.5 and has not yet measured Fable 5.2. In What common Claude Code tasks cost, measured on each model, five small tasks in a 1,176-line Python library averaged $0.110 per run on Opus 5.5 as Claude Code priced them on a Max plan on October 28, 2026, and all 15 runs passed their checks. Small, well-specified tasks like these already pass on Opus 5.5, so they are work to keep there. The study did not test migrations or long unattended runs, the work Anthropic's guide gives to Fable 5.2.

How does Fable 5.2 count against a Claude plan?

On Claude Max and Ultra, Fable models are included for up to 50% of your weekly usage limits, and on Claude Pro they run on usage credits billed on top of the plan, as of November 10, 2026 (Anthropic). Opus 5.5 is Claude Code's default model on every plan.

Claude Code pricing explained, with what a real task costs covers every plan's price and limits.

What breaks when you switch models mid-conversation?

Two things change when you switch between Claude Fable 5.2 and Claude Opus 5.5 mid-conversation, as of November 10, 2026: the new model has no prompt cache for the conversation, and a move down from Fable 5.2 loses its reasoning.

Anthropic's guide says to switch at a natural break: run /compact first in Claude Code, or start a new session with a short written plan, so the first request on the new model is small. On the API, Fable 5.2 also answers with a 400 error on every account when a request changes anything that came before one of its thinking blocks. Anthropic says nothing changes when Claude Code builds your requests.

How does AutoRouter choose between Fable 5.2 and Opus 5.5?

AutoRouter, the Mac app for Codex and Claude Code, picks a model for each task while it is the model selected. Your rules go first. Otherwise AutoRouter works out how hard the task is and only considers models that clear the quality bar for it. Hard work goes to the most capable model your plans cover, and to credits only when none is strong enough. Everything else goes to the model that costs you least. A task that continues keeps its model or moves up. In Claude Code, Claude models run on your Claude plan.

New models show up in the picker right away. You can choose Fable 5.2 or Opus 5.5 there yourself, which sends every request straight to that model, or write a rule in plain English. AutoRouter's guide gives "Never use Fable for any tasks." as an example, and an avoid rule is always respected. Every request shows its model, what it cost and why. AutoRouter is in beta and sometimes misjudges how hard a task is; you can rate the pick. It earns 10% on paid models and nothing on your plans.

Questions

Is Claude Fable 5.2 better than Opus 5.5 at coding?

On Anthropic's hardest coding benchmarks, yes. In its November 9, 2026 announcement, Anthropic reports 49.1% for Fable 5.2 at xhigh effort against 45.5% for Opus 5.5 at max on FrontierCode 2.0, and 69.9% against 62.3% on FrontierSWE v2. Vals AI, which ran Fable 5.2 at high and lists Opus 5.5 at max, has the two tied on Terminal-Bench 4.0 at 65.15% and Opus 5.5 ahead on SWE-bench Multimodal, 70.21% to 62.29%, as of November 10, 2026. For everyday coding, Anthropic recommends starting on Opus 5.5.

Should you raise effort on Opus 5.5 or switch to Fable 5.2?

Raise effort first, up to high. Anthropic's guide gives the order: check that the model can verify its own work with a test or a build, raise effort one level, then move to the next model up at a lower effort than you used on the smaller one. It adds that more effort doesn't give a model knowledge or skill it lacks.

How do you select Fable 5.2 in Claude Code?

Run /model fable in a session, or start one with claude --model fable. The fable alias points to Fable 5.2 from Claude Code v2.1.321, so run claude update first on an older version (model configuration). Fable 5.2 starts at medium effort until you change it with /effort. How to change the model in Claude Code covers every setting.

How much cheaper is Fable 5.2 than Fable 5.1?

Fable 5.2 is 20% cheaper per token: $8 and $40 per million input and output tokens against $10 and $50 for Fable 5.1, with cache reads at $0.20 against $0.25, as of November 10, 2026 (pricing). Anthropic also says Fable 5.2 generates output up to 30% faster than Fable 5.1.

Sources