Anthropic AI Models Slash Costs by Up to 45% With Fable 5.1

Anthropic has released two new Anthropic AI models, Claude Fable 5.1 and Mythos 5.1, in a move that directly answers the loudest complaints customers have voiced about the company’s earlier releases: cost, data privacy, and safeguards that got in the way of legitimate work. The upgrade lands with pricing that Anthropic says cuts costs by roughly 25 percent for typical use and up to 45 percent for heavier agentic workloads, alongside new permissions that let one of the models hunt for software vulnerabilities for the first time.

Key takeaways

• Anthropic launched Claude Fable 5.1 and Mythos 5.1, positioning both as direct responses to customer complaints about price, data retention, and overly cautious safeguards.
• Fable 5.1 costs about 25 percent less than Fable 5 for typical tasks and up to 45 percent less for complex agentic work, thanks to cheaper pricing on cached data.
• Mythos 5.1 running on low reasoning matches the benchmark scores its predecessor achieved on Max reasoning, according to commentary from Lisan al Gaib.
• Enterprise Frontier Safeguards will let customers store data on their own cloud servers instead of Anthropic’s, rolling out later this fall.
• Fable 5.1 can now be used to identify software vulnerabilities, though tasks like penetration testing still go to Anthropic’s Opus models.

Anthropic Unveils Claude Fable 5.1 and Mythos 5.1

Anthropic built Fable 5.1 and Mythos 5.1 to fix three specific pain points customers had flagged: pricing, how long data gets retained, and safeguards that were sometimes too aggressive. Rather than presenting this as a routine model refresh, Anthropic frames it as a direct answer to feedback from paying users who felt earlier versions were too expensive, too cautious, or too opaque about where their data ended up.

Launch Details

The two Anthropic AI models arrive with distinct roles. Fable 5.1 is pitched as the workhorse for coding and agentic tasks, while Mythos 5.1 keeps its focus on the same restricted domains — including biology — that its predecessor was built to handle carefully. Anthropic says Fable 5.1 delivers stronger performance than Fable 5 while running noticeably cheaper, a combination that matters for any company running large volumes of AI-driven tasks.

Early User Feedback

Before the wider rollout, a handful of early adopters got hands-on time with Fable 5.1, and their reactions have been notably upbeat. Every CEO Dan Shipper called it “the strongest coding model we’ve used,” adding that it’s fast, token-efficient, and “actually speaks like a normal person.” Box CEO Aaron Levie offered a similar take, saying his company’s agent running on Fable 5.1 picked up on subtleties and ambiguities in data that the previous Fable 5 model had missed during the same test. Those kinds of endorsements from enterprise leaders carry weight, even if they reflect early access rather than broad market testing.

Pricing and Performance Improvements in Claude Fable 5.1

The headline for many businesses will be the price. Fable 5.1 costs around 25 percent less than Fable 5 for everyday use, and up to 45 percent less for complex agentic tasks — the kind of multi-step, tool-using workflows that have become increasingly common in enterprise AI deployments.

Cost Reduction Metrics

That savings comes from reduced pricing on cached data, meaning information that has already been processed and stored costs less to reuse. For companies running repetitive or high-volume agentic pipelines, this kind of caching discount can add up fast, especially at scale.

Performance and Safeguard Enhancements

Beyond price, Anthropic claims Fable 5.1 simply performs better than its predecessor. It also comes with what the company calls “more precise safeguards” — filters that are less likely to block basic biology questions than they were in Fable 5. That’s a meaningful shift for researchers, educators, or students who may have previously hit unnecessary guardrails when asking straightforward science questions. Why this matters: overly broad safeguards have been a recurring friction point for AI companies, and tightening the precision of those filters without loosening safety more broadly is a delicate balancing act that shapes how usable a model feels in practice.

Mythos 5.1 Model Updates and Benchmark Results

Mythos 5.1 takes a more conservative path than its sibling. It keeps the exact same biology restrictions as the previous Mythos model, signaling that Anthropic isn’t loosening its guardrails in that specific domain even as it does so elsewhere.

On the performance side, commentator Lisan al Gaib pointed out something notable in early benchmark comparisons: Mythos 5.1 running on low reasoning scores the same as its predecessor did when set to Max reasoning. If that holds up across broader testing, it suggests Anthropic squeezed more capability out of a lighter compute setting, which could translate into faster, cheaper deployments for tasks that don’t need the model’s full reasoning power.

Enhanced Privacy with Enterprise Frontier Safeguards

Data privacy has been another sticking point for enterprise customers weighing AI adoption, and Anthropic is addressing it with a new offering called Enterprise Frontier Safeguards.

Data Retention on Customer Cloud

The feature promises “complete privacy” by storing customer data on the customer’s own cloud servers rather than on Anthropic’s infrastructure. For regulated industries or companies wary of handing sensitive data to a third-party AI provider, that architecture shift could remove a real barrier to adoption.

Rollout Timeline

Anthropic says Enterprise Frontier Safeguards will start rolling out later this fall. The company hasn’t detailed a specific date beyond that window, so businesses interested in the feature will need to wait for further updates as the rollout approaches.

New Cybersecurity Applications and Task Allocation

Perhaps the most consequential update is on the cybersecurity front. Anthropic is now allowing Fable 5.1 to be used for identifying software vulnerabilities — a use case it previously restricted.

Software Vulnerability Detection

This opens the door for security teams to fold Fable 5.1 into their vulnerability-hunting workflows, a task that increasingly sits at the center of how AI companies think about model safety and real-world risk.

Retention of Certain Tasks by Opus Models

Even so, Anthropic isn’t handing over the keys entirely. Certain cybersecurity tasks — specifically penetration testing, exploit generation, and binary-based vulnerability scanning — remain assigned to Anthropic’s Opus models rather than Fable 5.1. That distinction matters: it shows Anthropic is drawing a line between identifying weaknesses and actively exploiting them, keeping the more sensitive offensive capabilities on a separate, presumably more tightly controlled model. Why this matters: as AI models grow more capable of finding and exploiting software flaws, the industry is being forced to decide which capabilities get distributed broadly and which stay locked down, a tension that’s likely to keep shaping how companies like Anthropic structure their model lineups going forward.

FAQ

What are the main improvements in Anthropic’s Claude Fable 5.1 compared to Fable 5?

Claude Fable 5.1 offers about 25 percent lower typical cost, up to 45 percent cheaper pricing for complex agentic tasks, stronger performance, and more precise safeguards that less frequently block basic biology questions.

How does Mythos 5.1 compare to its previous version regarding biology restrictions and reasoning performance?

Mythos 5.1 maintains the same biology restrictions as the previous model and achieves similar benchmark scores with low reasoning settings as its predecessor did with maximum reasoning.

What privacy improvements come with Anthropic’s Enterprise Frontier Safeguards?

Enterprise Frontier Safeguards enhance privacy by storing data on customers’ cloud servers instead of Anthropic’s, and will start rolling out later this fall.

Can Claude Fable 5.1 be used for cybersecurity tasks like software vulnerability identification?

Yes, Anthropic now allows Fable 5.1 to be used for identifying software vulnerabilities, but certain tasks like penetration testing remain assigned to Opus models.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.