Anthropic rolls out Claude Fable 5, but it's available for a limited time
Anthropic is rolling out Claude Fable 5 with new guardrails and limited-time free access. BleepingComputer says it can burn through tokens much faster than Opus.
Intelligence analysis by GPT-5.4 Mini
Anthropic is trying to make a powerful model safer to release by adding guardrails and routing sensitive requests away from the new Fable 5 model. The tradeoff is cost: BleepingComputer says the model is extremely token-hungry, and Anthropic is only offering temporary free access before switching to usage-based pricing.
Anthropic made a very smart robot helper and added locks on the doors so it cannot easily help with dangerous requests. But this helper eats through its fuel very fast, like a car that burns a full tank in minutes.
Analysis
What Anthropic is doing
Anthropic has started rolling out a new model called Fable 5, described in the article as being built on the same underlying model class as Mythos, the company’s most powerful AI family. The company’s earlier Mythos announcement framed that model as potentially risky because it could help attackers target public and private software.
Safety controls and access limits
The article says Anthropic now believes it has added strong guardrails around the same model class. In Fable 5, sensitive requests involving areas like offensive cybersecurity, biology, or chemistry are supposed to be blocked or redirected to a previous model, Opus 4.8. The unrestricted Mythos 5 version is kept behind a tighter trust gate and is only available to vetted partners such as government cyberdefenders and specific life sciences researchers.
Cost and usage concerns
Anthropic is making Fable 5 free only for a limited window, through June 22, for Pro, Max, and Enterprise customers. After that, the company says it will move to usage-based pricing. BleepingComputer reports that the model consumes tokens quickly, especially in Workflow mode with high reasoning settings, with one test using roughly 1 million tokens in about eight minutes and exhausting a $100 Max plan’s daily usage in nine minutes.
Security angle
The article’s core tension is familiar: the safer a model becomes, the more expensive and constrained it may be to run. Anthropic appears to be testing how much capability it can expose broadly while still keeping higher-risk behavior behind controls. For security users, that means a more controlled front door, but also a model whose practical use may be limited by cost and rate of consumption.
Key points
- Anthropic is rolling out Claude Fable 5 as a safer version of its most powerful model class.
- Sensitive queries are supposed to be blocked or diverted to Opus 4.8.
- The unrestricted Mythos 5 version is reserved for highly vetted trusted partners.
- Fable 5 is free only until June 22, after which Anthropic says it will move to usage-based pricing.
- BleepingComputer reports that the model consumes tokens very quickly, especially in Workflow mode with high thinking settings.
If Anthropic’s guardrails work as intended, Fable 5 could give more users access to a powerful model without opening the door to easy abuse. That would make it easier for defenders to use advanced AI while keeping the riskiest requests routed to tighter controls.
If the token usage reported by BleepingComputer holds up broadly, the model may be too expensive for routine use outside short tests. The limited access model could also slow adoption if customers find the guardrails or pricing too restrictive.
