The news
On September 1, 2026, Anthropic released Claude Fable 5.1, its generally available frontier model, together with Claude Mythos 5.1, a version with looser safeguards for vetted users, and cut Fable's cache-read price by 75% to $0.25 per million tokens.
Anthropic said Fable 5.1 and Mythos 5.1 are the same model with different levels of safeguards. Fable 5.1 went on sale to all customers through the Claude API as claude-fable-5-1, Claude.ai and Claude Code, and on Amazon Web Services (AMZN), Google Cloud (GOOGL) and Microsoft Azure (MSFT). Mythos 5.1 is limited to U.S. organizations in Anthropic's trusted-access programs for cybersecurity and life sciences.
Headline rates did not change: $10 per million input tokens and $50 per million output tokens, the same as Fable 5. A token is a small chunk of text that AI providers bill by. Cache reads, which cover prompt content the model has already processed and is sent again, such as long documents or codebases, fell 75%; VentureBeat put the prior price at $1.00. Anthropic said that, measured over four weeks of actual usage in August 2026, typical workloads cost around 25% less than on Fable 5, and complex coding and highly agentic tasks up to around 45% less.
Anthropic also announced Enterprise Frontier Safeguards (EFS). The company said it introduced 30-day data retention starting with Fable 5 so it could correlate misuse across sessions and accounts. EFS instead stores that activity data in the customer's own Amazon S3, Azure Blob Storage or Google Cloud Storage account under the customer's encryption keys; automated systems scan it and send flags directly to the customer, with no review by Anthropic employees required.
Anthropic said it built EFS with more than 100 customers in industries including financial services, healthcare and the public sector, would roll it out in phases starting later in the fall and would not charge for it, though customers pay their own cloud storage and transfer costs. Until then, eligible customers get zero data retention (ZDR), meaning no stored prompts or outputs, on Fable 5 and Fable 5.1. Wells Fargo's chief information security officer was among the customers quoted backing the design.
The company said Fable 5.1's cyber protections trigger around 60% fewer interventions per Claude Code session than Fable 5's earlier safeguards, VentureBeat reported. On the Terminal-Bench-Science benchmark, Anthropic reported 52.6% for Fable 5.1 against 24.7% for Fable 5. These are vendor-reported results that have not been independently verified.
The numbers
- Fable 5.1 cache-read price
- $0.25 per million tokens (down 75%)
- Input / output price (unchanged)
- $10 / $50 per million tokens
- Anthropic's estimated cost cut, typical workloads
- Around 25% vs. Fable 5
- Anthropic's estimated cost cut, agentic and complex coding
- Up to around 45%
- Customers involved in designing EFS
- More than 100
Why CEOs should care
For CFOs and AI platform owners, the savings depend on how much of the bill is cache reads. Coding agents and retrieval-heavy apps that resend large contexts gain the most; workloads with little reuse see small changes, because input and output prices held. Anthropic's 25% and 45% figures are its own averages from August usage, so price your own August invoices at the new cache rate before resetting budgets.
For CISOs in regulated industries, EFS targets a known sticking point: a vendor holding their prompts for 30 days. Wells Fargo's security chief, quoted by Anthropic, credited the split, with logs held under the bank's own keys, for letting its teams use frontier models safely. Security teams should ask who reviews the flags, how long logs must be kept, what storage and egress will cost, and who is accountable if a flag is ignored.
Boards and buyers should note the two-tier structure: the same model is sold with different safeguards depending on who the customer is. Teams in security research or life sciences that need fewer refusals should plan for verification and monitoring to get Mythos-level access; everyone else should test whether Fable 5.1's revised cyber safeguards reduce the interruptions they saw with Fable 5.
The bigger picture
Frontier model pricing is moving at the margins rather than the headline. By cutting only the cache-read rate, Anthropic lowered the cost of long agent sessions that repeatedly resend the same context, the work it says gains the most, while leaving list prices unchanged.
The release also shows access tiers becoming standard: the most capable settings of a model sit behind verification, while general customers get the same model with tighter guardrails. Anthropic's twist is pairing misuse monitoring with logs the customer hosts, which could become a reference point for other labs selling into banks, hospitals and government.
What happened next
On September 17, Anthropic opened its Life Sciences Verification Program, one of the routes to Mythos 5.1. On September 22, it launched Claude Opus 5.5, which it said performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5, and on September 28 it released Claude Sonnet 5.5.
Watch for the phased EFS rollout Anthropic promised for later in the fall, and for whether zero data retention for eligible customers ends once EFS is available to them.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- Archive story
- It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
