Claude Fable 5.2

Claude Fable 5.2 is Anthropic’s AI model for difficult coding, document analysis and research that takes many steps. It can use tools to continue a task, check intermediate results and revise its approach. Anthropic released it on 9 November 2026, alongside the separately controlled Mythos 5.2.

Anthropic recommends starting with Claude Opus 5.5 for everyday coding and knowledge work, then using Fable for harder jobs. Its specifications list text and image input, text output, a one-million-token context window and up to 128,000 output tokens. Tokens are the units used to measure the text a model processes.

Standard API prices are $8 input and $40 output

Anthropic’s price list gives these standard rates in US dollars per million tokens:

Input and output prices are 20% lower than Fable 5.1’s $10 and $50 rates. Caching lets an application reuse previously processed material at a lower cost. The Batch API offers a 50% discount on input and output for work that can run asynchronously.

Optional fast mode costs $16 input and $80 output per million tokens. Anthropic reports up to 2.5 times faster token generation in that mode. Claude subscriptions have their own usage allowances; the token prices above describe API billing.

Fable has broad access; Mythos uses verification programmes

Fable 5.2 is available through Claude, the Claude API and supported Amazon, Google and Microsoft cloud services. Developers use the model ID claude-fable-5-2; access in Claude apps follows the plan’s model entitlements and limits.

Anthropic says Fable and Mythos share the same underlying model. Fable’s cybersecurity and biology safeguards support its wider release. Organisations can apply for Mythos through Anthropic’s Cyber or Life Sciences Verification Program. A Mythos evaluation result therefore needs its own label when comparing the two products.

Anthropic reports gains in coding and research

In Anthropic’s 9 November evaluations, Fable 5.2 scored 49.1% on FrontierCode 2.0 at Xhigh effort, compared with Fable 5.1’s 32.4%. On Terminal-Bench-Science 0.1 it scored 74.6%, against 52.6% for Fable 5.1.

These are vendor-reported results under the stated test conditions. They are useful evidence for selecting a trial workload; a project’s own tests establish how reliably the model handles that work.

Developers should check history handling and data retention

The API documentation specifies always-on adaptive thinking, a default effort of high and a June 2026 knowledge cutoff. Anthropic’s launch announcement describes medium as the default in Claude apps and Claude Code, so the setting depends on the product being used.

The migration notes tighten checks on edits to earlier conversation turns. Developers should also check compatibility when passing thinking blocks between models, using older computer-use tools or configuring an advisor model.

Fable 5.2 requires 30-day data retention unless Anthropic expressly authorises a zero-data-retention arrangement. Check that requirement against the information an application will send, alongside its permissions, usage budget and review process.