Anthropic has released “Fable 5,” a secure, guardrail-equipped iteration of its powerful “Mythos” AI model class. Designed to prevent the exploitation of sensitive cybersecurity and biological research data, the model is currently available to Pro, Max, and Enterprise users. According to reports from BleepingComputer, the model’s high computational demands lead to rapid token consumption, particularly when utilizing automated “Workflow” execution systems.
Why Did Anthropic Restrict the Mythos Model?
Anthropic originally developed the Mythos model class as a state-of-the-art tool, but the company identified significant security risks during its internal testing. In April, Anthropic warned that the model’s power could allow bad actors to identify and exploit vulnerabilities in software, including widely used applications like Firefox. To mitigate these risks, the company initially restricted access to Mythos, limiting it to a vetted group of government cyberdefenders and life sciences researchers. By applying strict safeguards to the new Fable 5 version, Anthropic aims to block or divert high-risk queries—such as offensive cybersecurity or chemical synthesis—to its older Opus 4.8 model.
How Does Fable 5 Affect User Token Consumption?
Fable 5 is significantly more resource-intensive than previous iterations, often consuming tokens twice as fast as the Opus 4.8 model. Testing conducted by BleepingComputer revealed that when the model is paired with “Workflow”—a feature that breaks complex prompts into parallel subtasks—token depletion happens in minutes. In one instance, a $100 Max subscription’s daily usage was exhausted in just nine minutes of high-effort operation. While casual interaction with the model is less taxing, the high-compute requirement explains why Anthropic has initially limited its deployment compared to older, more stable models.
What Is the Future of AI Model Accessibility?
Anthropic’s strategy for Fable 5 suggests a move toward usage-based pricing models for high-compute AI. While the company is offering free access to Pro, Max, and Enterprise customers for a limited time, this is scheduled to transition to a pay-per-use structure after June 22. Industry observers note that Anthropic has a history of launching restricted or “nerfed” versions of models before eventually expanding capacity. Whether the company will eventually optimize Fable 5 for general, high-volume use remains to be seen as the compute costs for such high-end models remain substantial.
Frequently Asked Questions
Is Fable 5 safer than the original Mythos model?
Yes. According to Anthropic, Fable 5 includes strict guardrails that block or redirect sensitive cybersecurity, chemistry, or biology queries to the older Opus 4.8 model.
Why does Fable 5 consume tokens so quickly?
The model requires significant computational power to process complex tasks. BleepingComputer testing indicates that usage is particularly high when using “Workflow” to run parallel subagents.
Can I still access the unrestricted version of the model?
The unrestricted version, marketed as “Claude Mythos 5,” is reserved exclusively for a highly vetted group of trusted partners, including government agencies and specialized researchers.
Stay ahead of the latest AI security developments.
Subscribe to our newsletter for deep dives into how evolving AI models are changing the cybersecurity landscape.
