Claude Fable 5.2 Costs $8 and $40 per Million Tokens, Exactly Twice Opus 5.5. Anthropic Says It Is Cheaper per Task. Here Is the 20-Task Test I'd Run Before Switching.
Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, and the API price for Fable 5.2 is $8 per million input tokens and $40 per million output tokens. That is exactly double Opus 5.5 at $4 and $20. Anthropic's pitch is that Fable 5.2 at High effort scores about level with Opus 5.5 at Max effort for roughly half the cost per task. If that holds on your work, the doubled sticker price is a discount. If it does not, you just doubled your bill.
What is actually confirmed
The model ID is claude-fable-5-2. The context window is 1M tokens with up to 128K output, the same as Opus 5.5. It is available through the Claude API and the cloud marketplaces listed in Anthropic's docs, and fast mode runs at $16 and $80 per million tokens on the platforms that support it. Batch jobs get 50% off.
One detail from the docs that is easy to miss: the API default effort for Fable 5.2 is high, while Opus 5.5 defaults to medium. If you swap the model ID and change nothing else, you are comparing Fable at high effort against Opus at medium effort, and your cost per request will move for reasons that have nothing to do with the model's quality.
Mythos 5.2 is a different story. It is limited to vetted organizations in cybersecurity and life sciences, and Anthropic says it shares Fable's specifications and pricing. Unless you work in one of those fields and get verified, it is not something you can rent.
The half-cost claim, read carefully
Anthropic's launch post uses its FrontierCode 2.0 agentic coding benchmark. Fable 5.2 at High effort scored 45.3%, Opus 5.5's best score was 45.5%, and Fable at Xhigh reached 49.1%. Anthropic says the High-effort result came at roughly half the cost per task of Opus 5.5 at Max.
Notice what is being compared. The baseline is Opus at its most expensive setting. Many of us do not run Opus at Max for everyday work. If your current setup is Opus 5.5 at Medium, the half-cost comparison does not describe your situation, and Fable may well cost more for you. I have not seen independent numbers on this yet, and the benchmark is Anthropic's own.
The headline claim from the post is that a team of 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours with minimal human intervention. That is formalization, meaning a machine-checked version of a known proof, not a new proof, and it is also Anthropic's report. It tells you the model can hold a very long task together. It does not tell you what your invoice parser costs.
The 20-task test I would run
Per-token price and per-task price are different numbers, and you pay the second one. Here is a made-up example to show the gap. A task that reads 200K tokens and writes 20K costs $1.20 on Opus 5.5 ($0.80 input plus $0.40 output) and $2.40 on Fable 5.2. For Fable to win, it has to need less than half the tokens, or fix the failure that makes you rerun the task.
What I would do this week:
- Pull 20 real tasks from your own logs, mixed across easy, medium and hard. Not benchmark problems, your tasks.
- Run each on Opus 5.5 at the effort you use today, and on Fable 5.2 at High.
- Record cost per task, pass or fail by your own check, and number of retries.
- Compute cost per passing task, not cost per call. A cheap failure that you rerun is not cheap.
Twenty tasks is a small sample, so treat the result as a direction, not a verdict. It is still more than a benchmark chart tells you.
What I'd actually do
I would not swap the model ID globally. I would add Fable 5.2 as a second route for the tasks where Opus currently fails or needs retries, and leave routine work where it is. If the 20-task run shows Fable at High clearing your hardest tasks in one pass, that route earns its place even at double the token price. If the run shows no gap, I would save the money and revisit when independent evaluations land.
Where this could be wrong
Anthropic may be right and my caution may cost me a cheaper bill for a few weeks. If your current default is Opus at Max, the half-cost claim applies to you directly, and moving sooner is the better call. The other risk is cost I cannot see yet: Fable 5.2 may spend more thinking tokens on easy tasks than the benchmark suggests. That is exactly what the 20-task run would show.
Author
Lukas
@lukcombinator