Anthropic Releases Claude Fable 5, Its First Public Mythos-Class Model, With Sensitive Queries Routed to Opus 4.8

Overview

Anthropic on June 9 launched Claude Fable 5, which it describes as “a Mythos-class model that we’ve made safe for general use,” according to Anthropic. The release brings to the public the model family that began as the leaked, much-debated Claude Mythos — first surfaced through an Anthropic data leak in March and previewed to a restricted group ahead of the release of Claude Opus 4.7. TechCrunch describes Fable 5 as “the first publicly available version of its Mythos model”.

What We Know

Anthropic says Fable 5’s capabilities “exceed those of any model we’ve ever made generally available” and that the model is “state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas,” per the company’s announcement. The same announcement states that “Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude models.”

The defining feature of the public release is a safeguard layer. Anthropic says it launched the model “with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8,” adding that the safeguards “trigger, on average, in less than 5% of sessions,” according to Anthropic. Amazon Web Services, which lists the model on its cloud, describes the routing more concretely: “Harmful prompts related to cybersecurity, biology, chemistry, and health fall back to receive a response from Opus 4.8 instead,” per the AWS News Blog. AWS also notes a billing detail: “When a harmful prompt is routed to Opus 4.8 instead of Fable 5, you pay only Opus prices.”

Fable 5 and Mythos 5 are priced at “$10 per million input tokens and $50 per million output tokens,” the company says — a level TechCrunch characterizes as “double the price of Opus 4.8”.

On availability, Anthropic says Claude Fable 5 “is available everywhere today” and that “from today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost,” with developers able to call the claude-fable-5 model identifier via the Claude API, according to the announcement. MacRumors corroborates that the model “is included in Pro, Max, Team, and seat-based Enterprise plans from today until June 22.” The model also reached Amazon Bedrock the same day; AWS says it is “available today on Amazon Bedrock in the US East (N. Virginia) and Europe (Stockholm) Regions,” per its blog.

Alongside the public model, Anthropic is releasing a restricted variant. It says Claude Mythos 5 “will initially be deployed through Project Glasswing, in collaboration with the US government, as an upgrade to Claude Mythos Preview,” according to the company. AWS frames the two models as the same system with different limits: “The same model without these limits is Claude Mythos 5 and it will only be available to a small group of vetted customers,” per the AWS News Blog.

Anthropic published endorsements from launch partners. Cursor CEO and co-founder Michael Truell said Fable 5 “is the state of the art model on CursorBench” and that “it’s opened up a class of long-horizon problems that were out of reach for earlier models,” in the announcement. GitHub Chief Product Officer Mario Rodriguez called the model “a real step forward for the developers GitHub serves.”

What We Don’t Know

Anthropic’s benchmark and capability claims are the company’s own; independent third-party evaluations of Fable 5 were not part of the launch materials reviewed here. The precise list of topics that trigger the fallback to Opus 4.8, beyond the broad categories of cybersecurity, biology, chemistry, and health, has not been detailed publicly. It is also unclear how many organizations have been approved for Mythos 5 or what the full scope of the Project Glasswing collaboration with the US government entails.

Analysis

The release lands soon after Anthropic’s public warnings about the trajectory of frontier systems. TechCrunch notes that Anthropic released the model “days after warning AI is getting too dangerous,” reporting that the company “warned that systems are advancing so rapidly that they may soon achieve recursive self-improvement (RSI), autonomously improving themselves without human intervention,” in its coverage. The structure of the Fable 5 launch — a generally available model that quietly hands off its most sensitive prompts to a lower tier, paired with an unsafeguarded Mythos 5 reserved for vetted partners — reflects an attempt to ship Mythos-class capability to the broad market while keeping its most contested uses behind access controls.