Two models, one brain, different guardrails
Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, describing them as the same underlying model with different levels of safeguards. Fable 5.1 is generally available, while Mythos 5.1 is offered only through trusted access programmes, with protections designed to support work in cybersecurity and the life sciences. This two-tier idea builds on an approach Anthropic used earlier: with the previous generation, the company routed high-risk requests, in areas such as cybersecurity, biology, chemistry and model distillation, to a less capable model, Claude Opus 4.8, and notified users when it did so, according to VentureBeat's reporting on Fable 5.
The logic behind splitting the release is simple. Rather than giving every capability to everyone, or refusing sensitive questions outright, Anthropic keeps the stronger model available for everyday work while adding a safety layer for the general public. Mythos 5.1 lifts some of those restrictions for approved researchers working in sanctioned domains.
Lower cost and a new privacy option
Pricing is one of the clearest changes. Anthropic says Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads wherever usage is billed by token, thanks to reduced pricing on cache reads. For heavily agentic work, savings can reach up to roughly 45%.
On privacy, Anthropic is introducing Enterprise Frontier Safeguards (EFS), which it says gives customers complete privacy equivalent to a zero data retention policy. EFS stores data in cloud infrastructure controlled entirely by the customer rather than by Anthropic, and it is rolling out in phases beginning later in the autumn. Until EFS is available, eligible customers can use Fable 5.1 with zero data retention. Safeguards have also been tuned to reduce false positives: in cybersecurity, Anthropic says its newest safeguards block 60% fewer false positives than before, partly because Fable 5.1 can now be used to find software vulnerabilities — though not to build exploits for them.
Where the model performs
Anthropic positions Fable 5.1 as its most advanced model for coding and knowledge work. By default it uses High effort in Claude Code and Medium effort in Claude Cowork and on Claude.ai. Even at Low or Medium effort, Anthropic says it matches or beats Fable 5 at a much lower cost.
Some benchmark figures published by Anthropic show the gap. On Terminal-Bench-Science 0.1, a test of agentic scientific research, Fable 5.1 scored 52.6% against Fable 5's 24.7% and Opus 5's 29.0%. On AutomationBench, which covers business workflows, it reached 31.4% versus 17.1% for Fable 5. On Humanity's Last Exam with tools it scored 65.0%. These are Anthropic's own reported numbers, and the company notes that its production safeguards were enabled during testing, which likely lowered scores on some tasks.
A real example: finding a bug nobody could explain
The most concrete illustration comes from the investment firm Millennium. According to Anthropic, Fable 5.1 found the cause of a rare crash in Millennium's internal systems that no engineer — and no other model — had explained after years of trying. Damien, a Senior Portfolio Manager at Millennium, described the case: "A particular piece of code had an extremely rare crash, about one in a million runs, that nobody on our team had explained in four to five years. Every model I tried, including Fable 5, missed it. Claude Fable 5.1 was the first to find it. It disassembled an external vendor library, matched it against the core dump, and traced the crash to a bug in that library."
Here is how a team could apply that pattern in practice. Suppose you have an intermittent production crash. You would give the model the core dump, the relevant logs, and access to the codebase, then prompt it with something direct: "Analyse this core dump against our service code and the linked vendor library. Identify the most likely cause of this one-in-a-million crash and explain your reasoning step by step." The value is not magic — it is the model's willingness to run a long, patient investigation across code and binaries that a human might never have time to justify.
What early users say
Several partners have shared results. Cognition co-founder Walden Yan said the firm was moving its Opus 5 traffic in Devin to Fable 5.1 on launch day, noting the new cache read pricing finally made a Fable-class model economical for workloads it had kept on Opus. Dan Shipper of Every summed up the appeal bluntly: "It's friendly Fable. Fable-level intelligence, Opus-level price, Sonnet-speed." MongoDB's Ron Sanzone described the model building a complex prototype over roughly three days, running unattended for hours with verification loops in place.
The takeaway
For teams doing coding and research work, the headline changes are practical rather than flashy: lower token costs, a clearer privacy option through EFS, and a safety model that routes risky requests instead of simply blocking them. Whether Fable 5.1 fits your budget depends on your workload, but the pricing shift makes it worth a fresh look.
Brain.mt can help you using AI for your business. Contact me for more information. I also offer dedicated workshops and training about this subject.
