Date: September 1, 2026 Main source: Anthropic Link: https://www.anthropic.com/claude-fable-and-mythos-5-1 Additional sources: https://techcrunch.com/2026/09/01/anthropics-new-fable-release-is-cheaper-less-restrictive/
Summary
Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 as its most advanced models for programming and knowledge work. Fable 5.1 is generally available, while Mythos 5.1 keeps restricted access for certain cybersecurity and life sciences work. The company also significantly lowered the cost of cache reads, an important variable for agents that reuse large contexts. Anthropic paired the launch with changes to enterprise safeguards and options for organizations that need more control over their data.
Why it matters
The effective cost of running agents is starting to be as strategic as the model's raw performance. Lowering the cost of reusing context favors long workflows, programming tools, research and enterprise agents that repeatedly query the same set of instructions or documents. The split between Fable and Mythos also shows how labs are creating distinct access tiers based on the risk of the capabilities.
Practical implications
- Teams with long-running agents should measure how much of their usage can benefit from prompt caching.
- It makes sense to evaluate Fable 5.1 against existing models using real tasks, not just public benchmarks.
- Companies with sensitive data should review the new retention options and safeguards before switching providers.