Claude Fable 5.1: is Anthropic's most expensive model worth it? Our test
Claude Fable 5.1 costs five times as much as Sonnet. On our three coding jobs it was right every time, but so were Sonnet, Opus and GPT-6.1 Sol, for much less.
Claude Fable 5.1 is Anthropic's most powerful AI model. It is also its most expensive: five times the price of Claude Sonnet 5.5. So we asked a simple question: on everyday coding jobs, is it worth it? We gave Fable 5.1 three small coding jobs, next to Opus, Sonnet, Haiku and GPT-6.1 Sol. Fable got everything right. But so did Sonnet, Opus and GPT-6.1 Sol, for a fraction of the price: one job cost 14.2 cents with Fable and 2.6 cents with Sonnet.
What is Claude Fable 5.1?
Anthropic launched Fable 5.1 in September 2026 and calls it one of "the world's most advanced models for coding and knowledge work", in its announcement. It is the successor of Fable 5.
There is a twin called Claude Mythos 5.1. It is the same model with different safety rules, and only a few vetted security and science teams can use it. Fable 5.1 is the version everyone can use.
Anthropic says Fable 5.1 is best at long, hard tasks: work that takes hours or days, where it has to plan, check its own work and find the real cause of a problem.
What does it cost?
If you pay per use, through Anthropic's API, Fable 5.1 costs $10 per million tokens read and $50 per million written. That is the same as Fable 5. An API is the connection that programs use to talk to Claude.
What did change is the cache. Programs like Claude Code keep a copy of earlier text, so the model can re-read it cheaply. Re-reading from the cache now costs $0.25 per million tokens, 75% less than before. Anthropic says this makes typical work about 25% cheaper than with Fable 5, and long, heavy work up to about 45% cheaper.
For comparison, from Anthropic's price list: Opus 5.5 costs $4 and $20, Sonnet 5.5 $2 and $10, and Haiku 4.5 $1 and $5.
Fable on a Max plan
If you have a Max plan ($100 or $200 a month), Fable 5.1 is included. You can spend up to half of your weekly limit on it, the help center says. It uses your limit up faster than other Claude models.
When you reach that half, you can keep going with extra paid credits, or switch to another model. The same rules apply to "premium" seats on Team plans.
Fable on a Pro plan
On a Pro plan ($20 a month), Fable 5.1 is not part of your limits at all. You can use it, but only with "usage credits": you pay for every use, at API prices. The same goes for "standard" seats on Team plans.
So on Pro, a long chat with Fable can cost real money on top of your subscription. Fable is not available on the free plan.
Our test: three small coding jobs
We used our new set of secret tests. The tests in our public test kit are known now, so we keep these private. There are three jobs:
- D. Fix a money reader: code that turns "€ 1.234,56" into cents, with several bugs.
- E. Split a restaurant bill: shared dishes, a tip, and cents that must add up exactly.
- F. Read a CSV file: a spreadsheet file saved as text. Commas, line breaks and quotes can hide inside a field, and errors must name the right line.
After each answer we ran our secret checks: 28 for job D, 17 for E and 17 for F. For example: in the CSV text 2,"Bo, Jr.", the name must be "Bo, Jr.", not "Bo" and "Jr.".
Each model did each job three times. We sent each job as one request through OpenRouter, a service that gives access to many models, with the default settings.
The results: Fable is right, but so are cheaper models
All four big models passed every secret check, in every try:
- Claude Fable 5.1: 186 of 186. About 32 seconds per job, 14.2 cents per job.
- Claude Opus 5.5: 186 of 186. About 27 seconds, 5.9 cents per job.
- Claude Sonnet 5.5: 186 of 186. About 17 seconds, 2.6 cents per job.
- GPT-6.1 Sol: 186 of 186. About 24 seconds, 1.2 cents per job.
- Claude Haiku 4.5: 157 of 186. The fastest and cheapest, at 0.8 cents per job, but it slipped on the money job and the CSV job.
So on these jobs, Fable cost about 6 times as much as Sonnet and about 11 times as much as GPT-6.1 Sol, for the same result. Jobs D and F caught Haiku out in every try. On the CSV job it slipped on details such as a stray line-break character, header rows and line numbers.
Which Claude should you pick?
For small, clear coding jobs, our test says: Sonnet 5.5. It was right every time, fast, and about 6 times cheaper than Fable. Haiku is even cheaper and faster, but check its work.
Fable 5.1 is built for something else: long, hard work over hours or days. Anthropic and its partners give examples, like finding a crash that engineers could not explain for years. Our short test cannot check that. If that is the kind of work you do, Fable or Opus may be worth trying. Just watch the cost, especially on a Pro plan.
Want to see how this works in Claude Code? Read our Claude Code pricing test. For a cheaper, open alternative, see our Qwen 3.8 test.
How we checked
On 7 October 2026 we read Anthropic's announcement, its price list and the help article on Fable in Claude plans. The screenshots are our own captures of those pages, cropped only.
We ran the test on 6 and 7 October through OpenRouter, with default settings and three tries per job per model. Sonnet 5.5 and GPT-6.1 Sol did jobs D and E on 6 October, for our Qwen article, and job F on 7 October. The costs are what OpenRouter charged. The owner of this site paid for the requests, about $2 in total. The charts were made by us. None of the images was made by AI.
This was a small test with three short jobs, sent as one message each. It says nothing about long projects, which is where Anthropic says Fable is strongest. Prices can change, so check Anthropic's page before you choose.