Verified, then logged.
Sun 22 Nov 2026 · 00:20 KST / 15:20 UTC한국어 | ENSearch /

01 / AI · Claim vs. Check · Analysis

Claude Fable 5.2 lowers input and output prices

Lower prices than the previous version leave data retention exceptions subject to each service’s terms.

Image from Anthropic
Image: Anthropic

Anthropic announced Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens for developers.

The lower prices sit alongside separate conditions on data retention. Anthropic recommended starting with an existing model for routine coding and using the new model for complex work, including delegating simpler tasks to smaller models or advising them on harder problems.

South Korea is included in Anthropic’s list of countries where Claude is available.

That country listing is separate from the access conditions for Mythos 5.2: Anthropic said organisations must pass its verification programmes. Pricing comparisons and service terms therefore address different parts of access.

Input token price comparison

InputFable 5.110
Fable 5.28
Source platform.claude.com · as of 2026-11-09 · unit: USD per million input tokensChart: made by LOGIT

Prices against the previous version

Anthropic’s release notes list Fable 5.1 at $10 per million input tokens and $50 per million output tokens. Those rates provide the comparison with the new version’s input and output prices.

Anthropic’s announcement lists the optional fast mode at $16 per million input tokens and $80 per million output tokens. The company said this mode increases output token generation speed by up to 2.5 times.

Anthropic described Fable and Mythos as sharing the same underlying model. The distinction it identified is Fable’s safeguards against misuse in cybersecurity and biology.

Results for FrontierCode 2.0, Anthropic’s coding evaluation, are company-published figures. The published table specifies Xhigh reasoning effort for the Fable models; unless otherwise indicated, scores use each model’s highest available reasoning effort setting.

Data retention exceptions depend on the service

Vercel’s notice states that prompts and outputs are not used to train Claude and are retained for 30 days. With zero data retention enabled, its AI Gateway returns an error instead of sending the request to Fable 5.2.

AWS set out default requirements of retention for up to 30 days and human safety review. On Amazon Bedrock, Amazon personnel carry out that review within the AWS boundary.

AWS also announced a zero data retention exception for eligible customers’ internal use. That provides an exception to its default retention requirement.

Vercel’s published terms instead block requests when zero data retention is enabled, making the service’s terms material to whether a request can proceed.

AWS announced plans for safeguards that keep data in cloud infrastructure controlled by the customer. It did not disclose an availability date for those safeguards.