Anthropic Cuts Fable 5.1 Prices by Up to 45 Percent
The company also relaxes some safety filters, claims record scores on Terminal-Bench 4.0, and keeps Mythos 5.1 limited to a small set of enterprise partners.
Symbolic image: in a data center, a person seen from behind pulls a GPU server module out of a rack while status lights blink along the chassis.
Anthropic now charges up to 45 percent less for Fable 5.1 on complex agentic workloads and about 25 percent less when requests reuse cached data.
At a glance
- Fable 5.1 costs up to 45 percent less on complex agentic tasks.
- About 25 percent cheaper when a request hits cached data.
- Anthropic claims record scores on Terminal-Bench 4.0 and Humanity's Last Exam.
- Mythos 5.1 goes only to Project Glasswing customers: Amazon, Google, Microsoft.
- Neither report gives an absolute price per million tokens.
Anthropic has shipped Fable 5.1 with a price cut that lands hardest on agent workloads: complex agentic tasks cost up to 45 percent less, and requests that reuse cached data run about 25 percent cheaper. A second model, Mythos 5.1, arrives alongside it but only for a short list of enterprise customers.
Where the savings actually land
The discount is not uniform. It is largest on long multi-step runs where the model calls tools and chains intermediate results, which is also where token spend tends to pile up. Shorter calls benefit through caching instead, at roughly 25 percent.
What the available reports do not contain is a per-million-token price. Both figures are relative to Anthropic's previous rate, so a saving can be estimated but not reconciled against any published dollar amount.
Loosened filters, and the limits of that
Anthropic frames the change as sharpening its safety layer rather than removing it. Fable 5.1 answers basic biology questions less restrictively than before, and it is now permitted to help identify software vulnerabilities – a core task for security teams that also carries an obvious dual-use edge.
Mythos 5.1 does not get the same treatment; its restrictions in the biological domain stay in place. A package called Enterprise Frontier Safeguards is slated to roll out in fall 2026.
Benchmarks and access
Anthropic reports record results on Terminal-Bench 4.0, which measures command-line programming, and on Humanity's Last Exam, a general reasoning test. The company also credits the model with three new scientific results, among them a custom GPU optimization and a high-resolution map of Venus.
Fable 5.1 is available on all platforms now. Mythos 5.1 stays inside Project Glasswing, whose members include Amazon, Google and Microsoft.
What buyers should check
For teams already running agents in production, the pricing line is the part that moves budgets: cost is usually why long tool chains get cut short. Whether a given workload sees the full 45 percent depends on how much of it is genuinely agentic. Until a token-level price list is published, that number is a ceiling rather than a forecast.
FAQ
How much cheaper is Fable 5.1?
Up to 45 percent on complex agentic tasks and about 25 percent on requests that hit cached data, measured against Anthropic's previous rate.
Can anyone use Mythos 5.1?
No. It is limited to enterprises in Anthropic's Project Glasswing program, which includes Amazon, Google and Microsoft. No general release has been announced.
Which safety restrictions did Anthropic relax?
Fable 5.1 is less restrictive on basic biology questions and may now assist with identifying software vulnerabilities. Mythos 5.1 keeps its biological limits unchanged.
