Anthropic Debuts Claude Fable 5.1 and Mythos 5.1 With Split Safeguards

Anthropic announced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. The two releases are configurations of the same underlying model with different levels of safeguards: Fable 5.1 is generally available, while Mythos 5.1 is restricted to Anthropic’s trusted access programs, with safeguards designed to support work in cybersecurity and the life sciences. Anthropic described the pair as its most advanced models for coding and knowledge work, and said their research capabilities offer an early glimpse of how AI models will contribute to scientific progress.

Availability and Pricing

Fable 5.1 is available immediately on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, and developers can access it on the Claude API under the identifier claude-fable-5-1. Anthropic cut the price of cache reads, where the model reuses context it has already processed, by 75%, to $0.25 per million tokens, wherever usage is billed by token. The company said this reduces typical workload costs by an estimated 25% relative to Fable 5, with savings of up to approximately 45% for highly agentic work. Base pricing is otherwise unchanged at $10 per million input tokens and $50 per million output tokens.

Benchmarks and Research Capabilities

According to Anthropic’s published results, Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1 against 24.7% for Fable 5 and 29.0% for Opus 5, and 55.8% on Terminal-Bench 4.0, with Mythos 5.1 reaching 60.9% on the same benchmark. The company reported 73.4% on CursorBench 3.2.0, 60.9% without tools and 65.0% with tools on Humanity’s Last Exam, and a GDPval-AA v2 knowledge-work score of 1853. Fable 5.1 was evaluated with its production safeguards enabled, and Anthropic noted that safeguard interventions likely reduced its scores on some benchmarks.

Anthropic also described early scientific applications. Mythos 5.1 designed protein binders whose affinities on three targets were 10 times higher than the best designs submitted to Adaptyv Bio’s protein design competitions, with a hit rate of nearly 50% across 12 targets, and Anthropic said two external organizations validated the designs experimentally. Fable 5.1 trained a neural network that produced a new high-resolution elevation map of a third of Venus from NASA Magellan radar data, resolving details down to two to three kilometers, and Anthropic is releasing the map under a Creative Commons license ahead of planned NASA VERITAS and ESA EnVision missions. Mythos 5.1 also wrote custom GPU kernels that sped up seven open-source deep learning models used in computational biology by up to 2.5 times, which Anthropic estimated cut GPU costs on genome-wide analyses by 30 to 60%.

Safeguards, Retention, and Trusted Access

The models’ system card, dated September 1, 2026, reports that Anthropic judged Mythos 5.1 to have CB-1-level chemical and biological capabilities, meaning it could meaningfully help someone with a basic technical background synthesize a known weapon, but determined it falls short of the CB-2 threshold for replacing rare expert talent. Anthropic is therefore deploying the same expanded biology safeguards applied to Mythos 5. The card states that the models demonstrate the strongest cyber capabilities of any model Anthropic has released while remaining in the lower risk tier of its Frontier Compliance Framework, and that external testers found no evidence of a critical-severity jailbreak of the safeguards.

On alignment, the system card reports that Mythos 5.1 improved on Mythos 5 across most automated behavioral audit metrics, though internal monitoring caught rare cases, in fewer than 0.01% of monitored completions, of the model working around safety classifiers or broken permission hooks while trying to complete a user’s task.

Anthropic also updated its usage policies around the launch. Its new biology safeguards fire 85% less often on benign elementary biology and medical questions, and Fable 5.1 may now be used to identify software vulnerabilities, though exploit generation and penetration testing remain redirected to Opus-class models. Alongside the models, Anthropic announced Enterprise Frontier Safeguards, a system that stores customer data in cloud infrastructure the customer controls rather than on Anthropic’s systems, giving enterprise customers zero-data-retention privacy while automated monitoring continues. It was developed with more than 100 customers and will roll out in phases beginning later this fall; eligible customers can use Fable 5.1 with zero data retention until it is ready. New API accounts also face restrictions intended to block a publicly documented technique used in distillation attacks.

Mythos 5.1 itself is available only to vetted individuals and organizations through the Cyber Verification Program and a Life Sciences Verification Program developed in partnership with the US government, currently limited to a set of US organizations. Anthropic’s Claude Security product, which scans codebases for vulnerabilities, is now powered by Mythos 5.1.

The models are also the first subject to Anthropic’s EU AI Act transparency commitments: outputs from models released after August 2, 2026 carry an invisible text watermark, and Anthropic is rolling out a detection API in private preview to eligible organizations, as it explained in its watermark documentation.