The Pulse
Anthropic launches Claude Fable 5.1 and Mythos 5.1
Anthropic has released Claude Fable 5.1 for general use and Claude Mythos 5.1 for vetted cybersecurity and life-science organizations. The models share the same underlying system, while pricing, safeguards and research capabilities mark the
AI.info Team ·
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, presenting them as the same underlying model configured with different safeguards. Fable 5.1 is generally available for coding, knowledge work and long-running agent tasks, while Mythos 5.1 is restricted to vetted organizations working in cybersecurity and the life sciences.
The launch combines a model upgrade with lower API costs and a change in how Anthropic handles sensitive workloads. The company says cache-read pricing falls 75% to $0.25 per million tokens, reducing typical workload costs by about 25% compared with Fable 5 and cutting highly agentic workloads by as much as 45%.
Anthropic also says Fable 5.1 can identify software vulnerabilities without developing exploits, while its revised cybersecurity safeguards produce 60% fewer false positives than the previous system. The company plans to offer its Enterprise Frontier Safeguards system to enterprise customers in phases beginning later this fall.
Fable 5.1 targets long-running software work
Fable 5.1 is available through Claude’s products, the Claude API and cloud platforms including Amazon Web Services, Google Cloud and Microsoft Azure. Developers can call it with the model identifier claude-fable-5-1. Anthropic lists pricing of $10 per million input tokens and $50 per million output tokens, with cache reads charged separately at the reduced rate.
The model is designed for tasks that run across multiple applications and require repeated verification. Anthropic says it can operate browsers, maintain records, recover after failed steps and continue unattended work through tools such as Claude Code and Claude Cowork. Its default effort level is high in Claude Code and medium in Claude Cowork and Claude.ai.
Early-access customers describe the main improvement as sustained performance rather than a single benchmark jump. “Claude Fable 5.1 is more comfortable with long, unattended work than Fable 5,” said Ben Lafferty, senior staff engineer at Shopify. “I’ve had workflows run for a long stretch without losing the plot: it keeps its own records, reprioritizes as things change, and picks up where it left off.”
Anthropic reports gains across coding and computer use
Anthropic’s published results show Fable 5.1 scoring 55.8% on Terminal-Bench 4.0, compared with 42.0% for Fable 5 and 52.3% for Claude Opus 5. On CursorBench 3.2.0, the new model scores 73.4%, ahead of Fable 5 at 70.5% and Opus 5 at 70.0%.
On the company’s version of Terminal-Bench-Science 0.1, Fable 5.1 records 52.6%, compared with 24.7% for Fable 5 and 29.0% for Opus 5. Anthropic cautions that the benchmark has a standard error of roughly 3.5 to 4.5 percentage points per model, and that its reproduction of earlier scores differs from the public leaderboard.
The company reports a 77.9% partial score and a 41.7% strict score on OSWorld 2.0. Anthropic says production safeguards affected some tasks, with restricted cybersecurity and biology work routed to other Claude models during testing. That qualification makes the results difficult to read as a pure comparison of model capability, but it also shows how safety controls remain part of the product’s performance profile.
Mythos 5.1 carries the higher-risk research work
Mythos 5.1 uses the same model as Fable 5.1 but removes or adjusts safeguards for approved work in fields where more capable assistance can create greater risks. Access currently goes to a set of U.S. organizations, with Anthropic coordinating with the U.S. government on broader access for domestic and international partners.
Anthropic says Mythos 5.1 produced high-affinity protein binders in testing conducted with open-source protein-design and folding tools. Across 12 targets, nearly half of the designs were viable binders, while the company says three targets produced binding affinities ten times higher than the best designs in Adaptyv Bio protein-design competitions.
The company also says Fable 5.1 trained a neural network to produce a higher-resolution elevation map covering one-third of Venus. The map improves detail from roughly 10–20 kilometers to two–three kilometers and estimates heights with up to 25% greater accuracy, according to Anthropic.
Lower prices come with a new privacy system
Anthropic’s Enterprise Frontier Safeguards system is intended to let customers keep data in cloud infrastructure controlled entirely by the customer rather than by Anthropic. The company describes the arrangement as providing privacy equivalent to zero data retention while retaining protections against adversarial use.
EFS is not available at launch. Anthropic says eligible customers can use Fable 5.1 with zero data retention until the enterprise system begins rolling out later this fall. The staged release places the privacy commitment alongside the model’s broader availability rather than making it a prerequisite for access.
Mythos 5.1 is subject to a separate access model. Cybersecurity users can seek entry through Anthropic’s Cyber Verification Program, while the company says enrollment for scientists in its biology program is expected to open soon.
Anthropic’s research claims still need outside validation
The release frames Fable 5.1 and Mythos 5.1 as tools for scientific work as well as commercial software development. Anthropic’s examples are specific: protein-binding experiments, GPU-kernel optimization for biological models and a Venus elevation map based on NASA Magellan radar data.
Anthropic says Mythos 5.1 sped up seven open-source deep-learning models by as much as 2.5 times through custom GPU kernels and cached intermediate results. The company estimates those optimizations could cut GPU costs for genome-scale analyses by 30% to 60%, and says it plans to open-source them.
The release does not establish that the models independently produced publishable scientific discoveries. It does show Anthropic moving from demonstrations of text and code generation toward systems that operate tools, run experiments and produce research artifacts under controlled access. For general users, Fable 5.1 is available now; the more permissive Mythos 5.1 remains limited to approved organizations.