Claude Fable 5.1 Is 25% Cheaper — Unless You Max Out Effort

Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1 — point upgrades arriving two and a half months after the originals, Fortune noted — and for most teams the story is the invoice, not the benchmark chart. Fable 5.1 is generally available now through the Claude API as claude-fable-5-1 and on AWS, Google Cloud and Microsoft Azure, and Anthropic estimates it costs about 25% less than Fable 5 for typical token-billed workloads — up to roughly 45% less on heavily agentic runs. One of Anthropic’s own pre-release testers disputes that framing, and the disagreement is the most useful thing in this launch.

Same model, two names, one gate

Anthropic is blunt: “Claude Fable 5.1 and Claude Mythos 5.1 are the same model, but with different levels of safeguards.” Fable is generally available. Mythos ships only through trusted access programs — US organizations only for now, via the Cyber Verification Program and the Life Sciences Verification Program, the latter developed with the US government, The Decoder reported.

The gap between them is narrow. On Terminal-Bench 4.0, an agentic coding test, The Decoder reported Fable 5.1 scores 55.8% and Mythos 5.1 reaches 60.9%, against 42.0% for Fable 5 and 37.3% for GPT-5.6 Sol. TestingCatalog noted production safeguards were active during testing and could lower some scores.

Effort level matters more than usual: Anthropic says Fable 5.1 defaults to High effort in Claude Code and Medium in Claude Cowork and on Claude.ai. Inside the $20-a-month Claude Code subscription, you are on High by default — not the Low or Medium settings where the savings live.

The 25% discount lives entirely in cache reads

Anthropic did not touch its headline token prices. The Decoder reported input stays at $10 per million tokens and output at $50; what changed is cache reads, down from $1 to $0.25 per million — a 75% cut, as TestingCatalog put it. Cache reads are what you pay when the model re-reads context it has already processed, which is exactly what long autonomous runs with many tool calls do over and over. So savings track your cache-hit ratio, not a flat discount.

The Decoder reported that high cost was the biggest complaint about Fable 5 and likely held back enterprise adoption, with the squeeze building since Opus 5 launched in late July at $5 input and $25 output per million — half of Fable’s rate.

An early tester says max effort costs more, not less

Artificial Analysis, which helped Anthropic with pre-release testing, disputes the savings claim, according to The Decoder. The cache-read cut saves about $1.40 per task on agentic workloads, it found, but Fable 5.1 at max effort costs 20% more per task than Fable 5, because it burns roughly 1.7 times as many output tokens: $3.76 per Intelligence Index task against $2.34 for Opus 5, which scores only three points lower.

Both claims can hold at once. Anthropic is quoting the price of a token; Artificial Analysis is counting how many tokens a finished task consumes. Anthropic’s own counter is the effort dial — it says that at Low or Medium effort, Fable 5.1 matches or beats Fable 5’s results at much lower cost.

Where your data goes now

Data retention was one of three areas of customer feedback Anthropic set out to answer, alongside price and safeguards. Its fix is Enterprise Frontier Safeguards, or EFS, which Anthropic says gives customers “complete privacy (the same as a zero data retention policy)” by storing data in cloud infrastructure the customer controls rather than Anthropic. It arrives in phases beginning later this fall.

The fine print is in the AWS post. AWS says Anthropic designated Fable 5.1 a Covered Model, a category carrying extra retention, safety-review and access policies. On Amazon Bedrock that means retention for up to 30 days and human review by Amazon personnel, and you must set the aws_review mode before invoking the model — though AWS says those prompts and outputs stay inside the AWS boundary and that Fable 5.1 “does not require sharing with the model provider.”

EFS-eligible customers can skip that: AWS says they get zero data retention on Bedrock and Claude Platform on AWS for internal use through December 31, 2026, with later EFS phases keeping data under the customer’s own encryption keys and audit logging.

Looser filters, watermarks and a distillation clamp

Anthropic says its newest cybersecurity safeguards block 60% fewer false positives — measured, per TestingCatalog, as false-positive cyber interventions per Claude Code session. The Decoder reported biology filters now fire 85% less often on harmless questions about basic biology and medicine.

The capability change underneath matters more. Anthropic says Fable 5.1 can now be used to discover software vulnerabilities, though not to develop exploits for them; The Decoder reported penetration testing and exploit generation still route to the Opus models. That line is enforced by Anthropic’s safeguards, not by anything on your machine — a distinction we went through before with Claude Code’s auto mode.

Two changes matter to compliance teams. These are the first Claude models with built-in watermarks, The Decoder reported — an invisible numerical watermark aimed at EU AI Act transparency, per TestingCatalog, with a detection API in private preview. And new API accounts can no longer edit Claude’s prior context while keeping the thinking transcript, closing what Anthropic calls a documented distillation technique. Fortune quoted the company saying “greater autonomy comes with new risks.”

What to test before you switch

The showcase evidence for the capability jump comes from Anthropic’s own testing: the investment firm Millennium found Fable 5.1 explained a rare crash its engineers had chased for years; Fable 5.1 remapped a third of Venus from NASA Magellan radar data at two-to-three-kilometer resolution; Mythos 5.1 wrote GPU kernels that sped up seven open-source models by up to 2.5 times. None of that tells you how it behaves on your repository.

• Price per completed task at the effort level you actually run, not per million tokens.
• Start at Low or Medium effort, where Anthropic claims parity with Fable 5 at lower cost.
• Measure your cache-hit ratio first; the whole discount lands there.
• Bench Opus 5 at $5 input and $25 output per million before assuming Fable 5.1 wins on value.
• On Bedrock, set aws_review and confirm zero-data-retention eligibility before sending regulated data.
• Re-run prompts that used to trip the cyber or biology filters, and regression-test formatting: Anthropic’s Felix Rieseberg said 5.1 leans less on bullet points and bold text.

Choosing a chat assistant rather than an API? Our $20-tier comparison of ChatGPT, Claude and Gemini covers that side.

What happens next

Three dates belong on a calendar: EFS reaching enterprise customers in phases later this fall, zero data retention on AWS running through December 31, 2026, and enrollment opening soon for scientists seeking Mythos 5.1’s biology capabilities, with international access planned.

Whether the benchmark jump survives production is the open question; The Decoder wrote that will become clear over the coming weeks. Fortune reported OpenAI’s next model, Astra, is due “soon,” with its most advanced cybersecurity capabilities restricted at first to alpha testers.

FAQ

Is Claude Fable 5.1 actually cheaper than Fable 5? It depends how you run it. Anthropic estimates about 25% lower cost on typical workloads and up to 45% on highly agentic work, from cutting cache reads to $0.25 per million tokens. Artificial Analysis measured it at max effort costing 20% more per task.

What is the difference between Claude Fable 5.1 and Mythos 5.1? Anthropic says they are the same model with different levels of safeguards. Fable 5.1 is generally available; Mythos 5.1 is restricted to vetted US organizations, according to The Decoder.

Where can I use Claude Fable 5.1 today? Through the Claude API as claude-fable-5-1 and on AWS, Google Cloud and Microsoft Azure, according to The Decoder. AWS says it is live on Amazon Bedrock, in AWS GovCloud (US), and on Claude Platform on AWS in North America.

Does Anthropic keep my prompts when I use Fable 5.1? Anthropic says EFS will store data in infrastructure the customer controls, in phases starting later this fall, with zero data retention for eligible customers until then. On Bedrock, AWS says Fable 5.1 carries up to 30 days of retention and human review by Amazon personnel.