·3 min read
Claude Fable 5.2 Launch: API Pricing and Migration Changes

Anthropic released Claude Fable 5.2 on November 9, 2026, for demanding reasoning and coding projects that span multiple steps. Its Claude API model ID is claude-fable-5-2. Standard input and output rates are each 20% lower than Fable 5.1. Teams that rebuild conversation history on each turn should test that path before switching production traffic. Anthropic launch announcement
Fable 5.2 specs and API pricing
Fable 5.2 has a one-million-token context window, a maximum output of 128,000 tokens, and always-on adaptive thinking. A large context window can hold more project material. Measure how many files and turns the model actually needs to finish a task. Fable 5.2 model overview
ModePer million input tokensPer million output tokens
Fable 5.1 $10 $50
Fable 5.2 standard $8 $40
Fable 5.2 fast mode $16 $80
At those rates, 100,000 input tokens and 10,000 output tokens cost about $1.50 on Fable 5.1, $1.20 on Fable 5.2 standard, or $2.40 in fast mode. This calculation holds token usage constant and excludes caching and additional tool charges. Claude Platform release notes
Anthropic reports up to 30% faster output generation than 5.1 and up to 2.5 times standard generation speed with fast mode. If your workflow spends most of its time running tests or waiting for tools, measure total task time separately. Anthropic launch announcement
Where to try Fable 5.2 first
Anthropic recommends starting with Opus 5.5 for everyday coding and knowledge work. If your current model keeps missing dependencies in cross-module changes, replay a migration you have already completed. For example, ask each model to move a shared date format to ISO 8601, identify readers and writers, propose a compatibility plan, and add tests. Compare missed callers, test results, and time spent on manual rework. Anthropic launch announcement
Test conversation history before migrating from 5.1
Fable 5.2 validates conversation prefixes for all accounts. A request can return HTTP 400 by default if it resends a thinking block after altering a system prompt, tool definition, or message that preceded the block. This can happen when a framework summarizes or reorders history after a tool call and then reattaches the earlier block. Official migration guide
Start with a fresh session in staging to confirm a basic response. Then run a multi-turn workflow that includes a tool result and inspect whether each request preserves the earlier history. Test a rollback to 5.1 separately. Fable 5.1 cannot read thinking blocks produced by 5.2, so it may need to reason again even when it still has the text history. Official migration guide
Frequently asked questions
How do Fable 5.2 and Mythos 5.2 differ?
They share an underlying model. Fable is available for general product and developer use with stronger safeguards. Mythos is available to eligible organizations through Anthropic verification programs. Anthropic launch announcement
Can a workspace with zero data retention switch immediately?
Fable 5.2 generally requires 30-day data retention. Workspaces using zero data retention should confirm explicit authorization from Anthropic before switching. Fable 5.2 model overview
Should fast mode be the default?
Establish a standard-mode baseline first. Then compare a sample of tasks where output generation is the bottleneck. At the listed rates, fast mode costs twice as much per token as standard mode, so compare the actual time saved per completed task. Official migration guide