AI Pulse · Frontier Labs Watch

Anthropic has kept the headline price for Claude Fable 5.1 at $10 per million input tokens and $50 per million output tokens. That might look like no change. The meaningful move is in cache reads, which have dropped from $1.00 to $0.25 per million tokens.

Caching matters when an AI system repeatedly refers back to the same material, such as company policies, product catalogues, customer history, codebases, or long-running agent instructions. Anthropic estimates this change cuts typical workload costs by about 25%, with bigger savings for heavily agentic work where the model repeatedly reads shared context. Fable 5.1 also supports a 1 million token context window and up to 128,000 output tokens, so it is aimed at substantial, ongoing workflows rather than one-off chat prompts.

For operators, the point is not that a model suddenly became cheaper at the sticker price. It is that the economics of repeatable AI processes may improve without a visible price cut. If you are building internal assistants or automated workflows that revisit the same documents and instructions all day, cache pricing can matter more than the headline token rate. Fable 5.1 is already available in Claude Code v2.1.257 and AWS Bedrock. Mythos 5.1 remains restricted to Anthropic’s cybersecurity-focused trusted-access program, so it is not a general option for most businesses yet.

This is the kind of thing we build into an AI command centre, tracking which model and pricing changes actually affect the cost of running your workflows.