Anthropic Launches Fable 5.1 With 75% Cheaper Cache Reads and EU Watermarks
Anthropic released Claude Fable 5.1 and restricted Mythos 5.1 on September 1, keeping base pricing at $10/$50 but cutting cache reads 75% to $0.25 per million tokens, with invisible watermarks on every output.
Anthropic shipped Claude Fable 5.1 and its restricted sibling Mythos 5.1 on September 1, and while the base price holds at $10/$50 per million input/output tokens, the change most agentic-workload teams will feel is in the cache: cache reads drop 75%, from $1.00 to $0.25 per million tokens. Per MacRumors and community math, that nets out to roughly 25% total savings for token-billed agent runs that lean on long cached contexts, which is most of them.
The Pricing Detail That Rewards Agent Architecture
The cache math is worth reading closely because it changes optimization priorities. Cache writes now cost $12.50 (five-minute TTL) or $20 (one-hour) per million tokens against the $10 base input price, so writing cache is a premium; reading it back is an 87.5% discount. Workloads that repeatedly reference a large, stable context, the exact shape of long-running agents, codebase analysis, and document pipelines, get rewarded twice: once by the discount and once by the 512-token minimum staying unchanged. Teams that re-send fresh contexts every turn save nothing. The pricing is effectively a nudge toward architectures that persist and reuse context, which happens to be the architecture agents need anyway.
Mythos 5.1 Stays Gated, and Every Output Is Watermarked
The two releases are the same underlying model with different safeguard configurations: Fable 5.1 is generally available through the API, while Mythos 5.1 remains restricted to trusted partners in cybersecurity and life sciences, consistent with how the Mythos line has been handled since its UK cyber-capability warning. The other quiet rollout: every Fable 5.1 and Mythos 5.1 output carries Anthropic’s invisible statistical text watermark on all platforms, per the official documentation, with an EU AI Act watermark-detection API entering private preview. That makes Anthropic the first major lab to ship statutory watermarking at scale, days after the EU designated ChatGPT a VLOSE under the DSA, and it sets the compliance template every lab serving Europe now has to match.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
How to Automate Workflows with Claude Code Routines
Learn how to use Claude Code's new routines to schedule tasks, trigger API workflows, and automate GitHub PR reviews on cloud infrastructure.
Claude Code Retrospective Details 5x Drop in Session Costs
Anthropic's new technical retrospective reveals that prompt caching and prefix compaction act as strict architectural constraints for complex agentic workflows.
Sony Music and Warner Chappell Sue Anthropic Over Training Data
Sony Music Publishing, Warner Chappell, and other publishers filed a new copyright lawsuit against Anthropic and its co-founders on August 29, alleging systematic torrenting and scraping of copyrighted works to train Claude.
Anthropic Routes Mythos 5 Into Defensive Security Tools
Anthropic is deploying Claude Mythos 5 through Claude Security, partner tools, a $35 million OSS fund, and expanded access for vetted defenders.
Real-Time Audit Logging Arrives in Claude Code and Cowork
Anthropic has expanded its Compliance API to capture activity logs, workspace interactions, and terminal commands across Claude Cowork and Claude Code.