Anthropic Ships Sonnet 5 With Agentic Safety and Pricing Details
1 July 20265 min read
AI
Anthropic Launches Claude Fable 5 With New Safeguards for High-Risk Capabilities
Anthropic says its newest generally available model can handle longer, more complex work, while a less restricted version will remain limited to approved cybersecurity organizations.
Illustrative image. Cedar S. Insights uses editorial stock photography; images do not depict specific events described in articles.
Anthropic released Claude Fable 5 on Tuesday, presenting it as its most capable model available to the general public. At the same time, the company introduced Claude Mythos 5, a version of the same underlying model that will initially be restricted to selected cybersecurity partners.
The paired release reflects a growing problem for frontier AI developers: the capabilities that make a model useful for software engineering, scientific research and complex analysis can also make it more valuable for cyberattacks or other harmful activity.
Anthropic says Fable 5 uses additional safeguards for sensitive requests. When its classifiers identify certain high-risk activity, the system may route the request to Claude Opus 4.8 instead. The company acknowledges that this conservative approach will sometimes block benign work, estimating that the safeguards will activate in fewer than 5% of sessions on average.
Mythos 5 removes some of those restrictions, but it is not being released broadly. Access is initially limited to organizations participating in Project Glasswing, Anthropic's programme for cyber defenders and critical infrastructure providers. The company also plans a separate trusted-access programme for a small group of life-sciences researchers.
Anthropic reported substantial gains in coding, visual analysis, knowledge work and long-running autonomous tasks. It also described internal experiments in protein design and molecular biology. These results should be read as company-reported findings rather than independent scientific validation. Anthropic says it intends to publish more of the research later.
The model costs $10 per million input tokens and $50 per million output tokens through the Claude API. Anthropic says Fable 5 is immediately available through consumption-based API and enterprise plans. Subscription access is being introduced in stages because demand and available computing capacity remain difficult to predict.
The launch points to a wider change in AI product design. The industry's most advanced systems are increasingly being presented not simply as chatbots, but as agents capable of carrying out extended projects, using tools and revising their own work. That raises the value of the systems, but it also makes access controls, monitoring and data-retention policies more consequential.
For business customers, one policy change deserves close attention. Anthropic says traffic involving Mythos-class models will be retained for 30 days for safety purposes, while not being used to train new models. Organizations considering sensitive legal, scientific or commercial work will need to examine how that retention requirement fits their own privacy and compliance obligations.
Why It Matters
The important story is not merely another benchmark increase. Anthropic is testing a tiered-access model in which the same underlying intelligence is offered with different safeguards depending on the user and use case. If that approach works, it could become a template for distributing increasingly capable AI systems without making every capability universally available.
Primary Sources
Our sourcing: Cedar S. Insights provides source-led editorial analysis. Reported company, institutional and regulatory claims are attributed to their original sources unless stated otherwise.
Corrections: If a material factual error is identified, Cedar S. Insights will update the relevant article and preserve the distinction between the corrected statement and supporting evidence.
Topics
AnthropicClaudeAI SafetyLarge Language ModelsCybersecurity