Fable 5.2 cut token prices, but heavy reasoning can raise the bill

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 as the same underlying model with different safeguards. Fable is available across platforms, while Mythos is limited to verified cybersecurity and biology organizations. Pricing fell to $8 and $40 per million input and output tokens.

Fable targets work above Opus, but Anthropic says start lower

Anthropic reports 49.1% on FrontierCode 2.0 at Xhigh and output up to 30% faster than version 5.1. It still recommends Opus 5.5 for daily work and Fable for open-ended, long-running tasks. Medium remains the default effort in Claude apps and Claude Code.

Token price stopped describing task cost

SemiAnalysis measured 16.4% on GDPx, while cost per task rose from $67.16 to $88.15. Fable tried hundreds or thousands of tests where older models stopped near 20 and launched supporting agents. Users pay less per unit but more for thoroughness.

Higher effort does not guarantee a higher score

In Anthropic’s FrontierCode chart, Fable reached 49.1% at Xhigh for $14.19, while Max fell to 45.6% for $44.70. One benchmark cannot select every workload, but it defeats the easy assumption that more inference compute always buys a better result.

Teams must measure completed work, not API prices

Useful metrics include cost per accepted patch, corrections after review, latency and success by task type. A sensible router will send routine work to cheaper models and call Fable only when expensive verification reduces human rework.

Lilith's verdict

A cheaper token is the price tag on one screw. The machine’s bill depends on how many Fable uses before the operator can press the green button.