Claude Fable 5.2: what it is, pricing and how to use it

Claude Fable 5.2 is Anthropic's new model for long, difficult technical work, released November 9, 2026. It costs $8 per million input tokens and $40 per million output tokens, and you can use it now on any paid Claude plan or through the API as claude-fable-5-2.

What Anthropic posted on X

@claudeai said the model is "available today on all paid plans," on the Claude Platform and the major clouds, at a price down from $10 and $50. That is 20 percent below Fable 5.1, which shipped in September. Anthropic's announcement also claims output up to 30 percent faster.

Mythos 5.2 is the same underlying model with fewer cybersecurity and biology safeguards, and it goes only to vetted organizations through programs like the life sciences verification track. Fable 5.2 still blocks most high-risk security work and most research biology, and hands those requests to an older Claude model. Anthropic reports about 55 percent fewer blocks than Fable 5.1.

Claude Fable 5.2 pricing and limits

Subscribers should read the launch promotion terms before getting attached. Pro plans and standard Team seats get Fable through November 16 at 11:59 pm PT, up to 50 percent of weekly limits, plus one free limit reset. After that week, the article says Fable usage on those plans moves to usage credits, if you have them turned on.

Max, Ultra and premium Team seats get one to three resets based on how long they've subscribed, good through March 8, 2027. You claim one under Settings > Usage, and your account needs to have been paying before launch.

How to use Claude Fable 5.2

In Claude Code, you need version 2.1.321 or later, and no plan defaults to Fable, so select it yourself:

/model fable
/effort high

On the API, the effort setting goes in output_config:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-fable-5-2",
    "max_tokens": 16000,
    "output_config": {"effort": "high"},
    "messages": [{"role": "user", "content": "Here is our quote follow-up process: ... Tell me what is wrong with it before you suggest anything."}]
  }'

That prompt follows a tip from Anthropic's Alex Albert (@alexalbert__), who says to ask what's wrong with your idea before it builds anything because it will push back. He also suggests telling it up front what needs your approval, like deleting things, and giving it a whole project instead of the next step.

Which effort level to pick

Effort decides how long the model works on a problem, and this is where the money goes. Fable 5.2 defaults to medium in the Claude apps and Claude Code, but high on the API, so the same prompt can behave differently depending on where you run it.

Albert's advice is "Start on med/high effort, and save xhigh for hardest problems." Anthropic's docs say to start at high and keep max for measuring the ceiling. Outside testing points the same way: @ValsAI ranked Fable 5.2 first on its index and found high to be the "most performant effort," ahead of low, xhigh and max.

Anthropic's own chart shows why high is the cheap choice. At that setting Fable 5.2 scored 45.3 percent on FrontierCode 2.0, about level with the 45.5 percent best from Opus 5.5, at roughly half the cost per task. At xhigh it reached 49.1 percent, the highest of any released model Anthropic has tested.

Is it worth it?

Anthropic itself recommends starting most coding and knowledge work on Opus 5.5 and moving up to Fable for ambitious, open-ended projects. I'd take that at face value. Its headline example, an operating system built from an empty folder by more than 1,000 subagents over eight days, is impressive and has little to do with answering your customers.

For automation work, look at AutomationBench, which appears in Anthropic's own table. Fable 5.2 scored 51.4 percent, up sharply from 31.4 percent for Fable 5.1 but behind Gemini 4 Carbon at 53.9 percent. Both still fail close to half of those business tasks. These are launch-day numbers, too, and most come from the company selling the model.

What this means for businesses using AI

Run a real job from your business at high effort, then at medium, and compare the output and the bill before you switch anything. Most routine steps won't need Fable, and any step that can delete or send something should wait for a person's approval. If you want help sorting which of your workflows justify a top-tier model, New Face Design's free process audit starts there.