Anthropic's public Fable 5 model lacks the autonomous vulnerability discovery that made its preview famous. A fallback mechanism routes cyber queries to an older model. Pricing is $10/$50 per million tokens, but efficiency gains may offset the cost.
Anthropic released a public version of its Mythos AI model on Tuesday. The new Claude Fable 5 is the startup's most powerful model for general use. It also lacks the cybersecurity capability that made the preview version a global story.
The preview Mythos model could autonomously find and exploit software vulnerabilities. Anthropic showed that capability to about 200 organizations, including the U.S. government, under the Glasswing program. The company said in April that Mythos had uncovered thousands of software vulnerabilities.
Fable 5 does not have that capability. Anthropic built a fallback mechanism instead. When a user asks for help finding vulnerabilities in a specific code package, Fable 5 refuses the request and routes the query to Opus 4.8, an older model without the offensive cyber capability.
"Let's say I'm a college student asking the model like help me find cyber vulnerabilities on X package or code. The model would refuse and Fable 5 will fall back to Opus 4.8 for a response," Dianne Penn, Anthropic's head of product management, research and labs, told Reuters.
The core difference between the preview and the public version is simple. The preview Mythos model could autonomously find and exploit software vulnerabilities. The public Fable 5 cannot.
Anthropic tested extensively to ensure users could not jailbreak the system and force Fable 5 to perform restricted actions. The fallback mechanism is a safety layer that prevents the model from being used as a weapon.
For developers and security researchers who relied on the preview, the public version is a downgrade in one specific dimension. They lose the ability to automate vulnerability discovery at scale.
For everyone else, the fallback is invisible. The model still performs software engineering, analytics, and general reasoning tasks at the same level as the preview. The only blocked domain is offensive cybersecurity.
Anthropic said users who had access to the preview version of Claude Mythos, the version without guardrails, would be able to upgrade to the new Claude Mythos 5. That suggests the company is maintaining two tracks: a restricted-access version for vetted organizations and a public version with guardrails.
Anthropic set pricing at $10 per million input tokens and $50 per million output tokens for both Fable 5 and Mythos 5. That is higher than the $3 per million input tokens and $15 per million output tokens for Opus 4.8.
Penn said early customer feedback showed that Fable 5 accomplishes tasks with lower token usage, bringing the overall cost per task down. The model is more efficient per token. Even at a higher per-token price, the total cost for a given task may be lower.
The math works like this. If Opus 4.8 uses 1,000 tokens to complete a task at $15 per million output tokens, the cost is $0.015. If Fable 5 uses 300 tokens for the same task at $50 per million output tokens, the cost is $0.015. Same cost, better performance.
For tasks where Fable 5 uses significantly fewer tokens, the cost advantage flips in its favor. That is the bet Anthropic is making: the model's efficiency gains will offset the higher per-token price.
Anthropic's valuation has climbed to $965 billion, putting it ahead of OpenAI in the private market. Both companies are racing to go public. The Mythos rollout is a key part of that narrative.
OpenAI has not released a model with comparable autonomous vulnerability discovery capabilities. That gives Anthropic a differentiated product in the enterprise security market, even with the guardrails on the public version.
The restricted-access track under the Glasswing program gives Anthropic a direct line to government and defense customers. The U.S. government is one of about 200 organizations with access to the unguarded version. That relationship could become a recurring revenue stream as the government expands its use of AI for cybersecurity.
Google DeepMind has not publicly released a model with autonomous vulnerability discovery. Microsoft has integrated AI into its security products but has not offered a standalone model with offensive capabilities.
Anthropic's first-mover advantage in this niche is real. It is also narrow. The public version of Fable 5 does not have the capability that made the preview newsworthy. The restricted-access version is limited to 200 organizations. The addressable market for the unguarded model is small by design.
For traders tracking the AI sector, the Mythos rollout creates concrete markers.
What confirms the thesis:
What weakens the thesis:
Anthropic is making a calculated bet. It is sacrificing the most attention-grabbing capability of its model in exchange for regulatory and reputational safety. The bet is that the public version's efficiency gains and the restricted-access program's government contracts will generate enough revenue to justify the trade-off.
Practical rule: The public version of Fable 5 is a capable model with a specific capability removed. For most enterprise use cases, the removal does not matter. For cybersecurity firms and government agencies, the restricted-access version is the only one that matters.
The next marker is adoption data. If Anthropic reports strong token usage growth in its next public update, the bet is working. If adoption stalls, the guardrails may be costing more than they protect.
Prepared with AlphaScala research tooling and grounded in primary market data: live prices, fundamentals, SEC filings, hedge-fund holdings, and insider activity. Each story is checked against AlphaScala publishing rules before release. Educational coverage, not personalized advice.