#AnthropicReleasesFable5Model Anthropic Releases Claude Fable 5: The "Muzzled Genius" AI Now Available to the Public

Anthropic has officially released Claude Fable 5, its most powerful AI model for public use, ending months of speculation about the "too dangerous to release" Mythos-class technology. The launch, which took place on June 9, 2026, marks a significant shift in Anthropic's strategy, making frontier-level AI available to general users for the first time—albeit with significant safety restrictions .

While the underlying technology rivals the best in the world, the story of Fable 5 is as much about what it can't do as what it can.

The "Muzzled Genius"

To understand Fable 5, you have to understand its origins. In April, Anthropic unveiled Claude Mythos—a model so powerful at identifying software vulnerabilities that the company deemed it "too dangerous" for wide release, restricting it to a handful of cybersecurity partners .

Fable 5 is essentially the same model as Mythos, but with a critical difference: Guardrails. Anthropic has installed safety classifiers that act as a kill switch. If a user asks Fable 5 to help with high-risk tasks involving cybersecurity exploits, biology, chemistry, or model distillation, the system automatically hands the query off to the much weaker Claude Opus 4.8 .

Analogy: As one reviewer put it, it’s like being given the keys to a Ferrari, but the moment you try to press the accelerator too hard, the car swaps itself for an electric scooter .

Unmatched Performance (When Allowed)

Despite the "muzzle," Fable 5 is a beast. Anthropic states that for allowed tasks, it surpasses all previous public models in software engineering, knowledge work, and visual recognition .

· Code Migration: Stripe reported that Fable 5 completed a migration of a 50-million-line codebase in a single day—a task that would have taken a human team over two months .
· Gaming Logic: In a unique test of visual reasoning, Fable 5 successfully played and completed the Game Boy game Pokémon FireRed using only visual inputs. Earlier models required complex auxiliary equipment to play the game .
· Benchmarks: Fable 5 reportedly outscored competing models from OpenAI and Google DeepMind on nearly all industry benchmarks .

Pricing and Access: The Fine Print

While Fable 5 is "public," accessing it comes with a complex pricing structure that has frustrated some early users.

· API Cost: The official API rate is $10 per million input tokens** and **$50 per million output tokens—roughly double the price of Opus 4.8 .
· Subscription Confusion: Anthropic initially offered Fable 5 to Pro, Max, and Team subscribers for free until June 22, 2026. After that date, access will switch to a credits-based model, forcing users to pay extra even if they have a subscription, until server capacity catches up with demand .

The "Shadow" Model: Mythos 5

Alongside Fable, Anthropic released Claude Mythos 5 for approved partners. This is the unrestricted version of the model, with all safety classifiers removed. It is currently only available through Project Glasswing, a collaboration with the U.S. government and critical infrastructure providers .

Early results are staggering:

· Biotech: Internal experts using Mythos 5 sped up drug design processes by roughly 10 times. In protein design, 9 out of 14 targets produced strong candidates .
· Cybersecurity: In testing, Mythos 5 wrote an exploit for a known Windows kernel vulnerability in just 31 minutes .

A Necessary Precaution?

Anthropic's cautious release highlights growing tension in the AI industry between innovation and safety. The company admitted that Mythos 5 narrowly avoided crossing the "CB-2" red line, which would have meant the model could help create novel biological weapons rather than just identify existing ones .

Data Retention: In a controversial move for privacy advocates, Anthropic has imposed a mandatory 30-day data retention policy for all Mythos-class traffic, overriding previous "zero retention" enterprise agreements, citing the need to monitor for "novel jailbreaks" .

The Bottom Line

Claude Fable 5 represents a new plateau in AI capability that is now available to the public. However, it is a "productized" version of a revolutionary technology, heavily restricted and expensive to run.

For developers and enterprises, it is the most powerful tool on the market for long-horizon, complex tasks. For the average user, the free access window is short, and the "guardrails" are very real. Whether this "muzzled genius" model becomes the industry standard for frontier AI releases—balancing raw power with safety switches—remains to be seen.