Anthropic's Claude Fable 5.1 cuts agentic task costs by up to 45 percent

Anthropic has released two new AI models, Claude Fable 5.1 and Mythos 5.1, a refresh the company positions as a direct answer to customer criticism around pricing, data retention and overly aggressive safety filters, according to The Verge.

The headline number is cost. Anthropic says Fable 5.1 delivers stronger performance than Fable 5 while costing about 25 percent less in typical use, and up to 45 percent less for complex agentic tasks — the multi-step workloads where a model plans, calls tools and iterates across turns. The savings come from lower pricing on cached data, meaning input that has already been processed and stored, which disproportionately benefits agents that repeatedly re-read large amounts of context.

What early users are saying

Reactions from early-access customers appeared alongside the announcement. Every CEO Dan Shipper called it "the strongest coding model we've used," noting that it is now fast, token-efficient and, in his words, "actually speaks like a normal person." Box CEO Aaron Levie reported that his company's agent running Fable 5.1 picked up on subtleties and ambiguities in data that Fable 5 missed when given the same test.

The Verge also highlights a benchmark observation from Lisan al Gaib: Mythos 5.1 set to low reasoning scored the same as its predecessor configured at Max reasoning, implying users can reach equivalent results in cheaper settings.

Looser safeguards, new security permissions

Fable 5.1 introduces what Anthropic calls "more precise safeguards," which the company says are less likely to block basic biology questions than Fable 5 was. The change is not uniform across the pair: Mythos 5.1 keeps the same biology restrictions as the earlier model.

Security policy is shifting too. According to The Verge, Anthropic now allows Fable 5.1 to be used for identifying software vulnerabilities, but it will still redirect certain cybersecurity tasks — penetration testing, exploit generation and binary-based vulnerability scanning — to its Opus models.

Enterprise data retention

On privacy, Anthropic says its Enterprise Frontier Safeguards offer "complete privacy" by keeping customer data on the customer's own cloud servers rather than on Anthropic's infrastructure. The Verge reports that the feature will begin rolling out later this fall.

Availability

Claude Fable 5.1 is available now on all platforms. Mythos 5.1 is more limited, offered only to participants in Project Glasswing.

Why it matters

The release targets the three complaints enterprises raise most often about frontier models: cost, compliance and refusals. Agentic workloads are where inference bills compound, because agents re-read context and loop through tool calls, so a 45 percent cut driven by cache pricing changes the economics of deploying them at scale. Softening refusals on ordinary biology questions — while keeping the tighter limits on the Mythos line — suggests vendors are calibrating guardrails per model rather than applying them uniformly. And letting enterprise data stay on a customer's own servers removes a common procurement blocker. Competitors pitching many of the same enterprise customers now face a release that competes on cheaper agents, not just raw model quality.