Anthropic's launch figures for Claude Fable 5.2 show the effort setting moves both score and cost per task by large amounts. Medium is the default in the Claude apps and Claude Code, and Anthropic's own numbers say Low, Medium and High can each match Opus 5.5 on specific tests for a fraction of the cost.
Claude Fable 5.2 launched on November 9, and the most useful detail for anyone building on it is how much the effort setting changes the economics. Anthropic says Fable 5.2 runs at Medium effort by default in the Claude apps and Claude Code. If you are calling the model from your own pipeline, do not assume that default is the right one for every job. Anthropic's launch figures show the same model landing at very different cost and quality points depending on the dial.
Start with the cheap end. Anthropic says that on Humanity's Last Exam at Low effort, Fable 5.2 matches the top score of Claude Opus 5.5 for under one tenth of the cost per task. On the WANDR benchmark at Medium effort, Anthropic says Fable 5.2 matches Opus 5.5's top score at about 40 percent of the cost per task. And on its Chartography benchmark at Medium effort, Fable 5.2 scores 66.4 percent, versus 64.4 percent for Opus 5.5 at Max effort. For question answering, extraction and chart or document reading, those results suggest you should test Low and Medium before reaching higher.
Coding is where the upper settings earn their price. Anthropic says that at High effort on FrontierCode 2.0, Fable 5.2 scores 45.3 percent, about level with the best score of Opus 5.5 at 45.5 percent, at roughly half the cost per task. At Xhigh effort it reaches 49.1 percent, the highest score of any released model Anthropic tested. That last step is worth 3.8 points over High, so the question is whether those points are worth the extra spend on your task. For a one off migration or a hard bug that a person would otherwise burn a day on, probably yes. For a bulk code review that runs thousands of times, probably not.
One framing caution. These are Anthropic's own results on a chosen set of benchmarks, and they say nothing certain about your workload. Anthropic's launch page points everyday work toward Claude Opus 5.5, with Sonnet 5.5 and Haiku 5.5 below it, so Fable 5.2 is positioned for harder problems rather than as the default for everything. The right move is to measure rather than trust any scorecard, including this one.
Try it today
- Pick 20 to 50 real tasks from your own workload and save the inputs and the answers you consider correct.
- Run them on Claude Fable 5.2 at Low, Medium and High effort, and record score, latency and total cost per task for each.
- Choose the lowest setting whose score is within your tolerance of the best result, and set effort explicitly in your calls instead of relying on the default.
- Reserve Xhigh for a short list of hard tasks, such as difficult debugging or large refactors, and compare its result against High before you commit.
- Re run the same set against Claude Opus 5.5 so you know whether Fable 5.2 is worth its price for each task type.