Claude Fable 5.1: cheaper cache reads and integration changes

The September 1 release lists a default one-million-token context, up to 128,000 output tokens and always-on adaptive thinking. Official rates per million tokens are $10 input, $50 output and $0.25 for cache hits. A hit therefore costs 2.5% of base input, or one quarter of Fable 5’s previous cache-read rate.

Cache writes remain $12.50 for five-minute storage and $20 for one-hour storage. Total task cost also includes new input, writes, output and any other billed features. Reading one million already-cached tokens costs $0.25 for that read, not for the entire task.

The migration guide says forced tool selection with any or tool returns an error; auto and none remain available. Earlier models cannot reuse Fable 5.1 thinking blocks. Changes to preceding prompts, tools or messages can also invalidate later thinking blocks. Applications that assemble histories or switch models need to account for this.

Launch channels include Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. The two AWS options are distinct; platform and regional pricing also differ. The former AA ranking was a September 2 observation, not a current ranking, so it has been removed from the headline while the historical measurement link remains.