Anthropic releases Claude Fable 5, a guarded version of its Mythos model, to the public

Anthropic released Claude Fable 5 on Tuesday, the first model from its advanced Mythos class to be made widely available. The launch comes less than three months after the company restricted Mythos to a small group of partners over cybersecurity concerns, citing its ability to identify vulnerabilities in critical infrastructure.

What Fable 5 can do

Anthropic says Fable 5 exceeds every model it has previously made public, outperforming its own Opus 4.8 as well as GPT-5.5 and Gemini 3.1 Pro from rivals OpenAI and Google on internal benchmarks. The company highlights software engineering, knowledge work, vision, and scientific research as areas of strength. Ethan Mollick, an AI researcher and associate professor at Wharton, wrote that Fable consistently "outperformed basically every other public model I have used by a considerable margin" and could work for up to a dozen hours on multi-page specifications. In one demonstration, Mollick generated several video games, including a snake game and a subterranean exploration title, from a single prompt.

It was capable across many problems and produced some startling results -- it would work up to a dozen hours executing on multi-page specifications.

Safety guardrails

Fable 5 wraps the underlying Mythos capability in new safeguards. Requests involving cybersecurity, biology, chemistry, or model distillation are automatically routed to Claude Opus 4.8, a less capable model released in late May. Users are notified when a fallback occurs. Anthropic says more than 95% of Fable sessions run entirely on its own responses. The company also ran an external bug bounty and worked with red-teaming organizations, logging over 1,000 hours of testing without finding a universal jailbreak. As a further measure, Anthropic will require a 30-day retention period on all traffic, even for enterprises that previously had zero-retention agreements, though it says the data will not be used for training.

Anthropic is being deliberately more conservative at launch.

The unrestricted version

Alongside Fable 5, Anthropic is deploying Claude Mythos 5 to organizations already approved for Mythos access, including roughly 200 partners in more than 15 countries enrolled in its Project Glasswing program. Mythos 5 lifts certain safeguards for approved cybersecurity and biology researchers and is being deployed in collaboration with the US government. The White House has established an arrangement to test the most powerful models from leading AI companies before release.

Pricing and access

Fable 5 is available through the Claude API and consumption-based Enterprise plans at $10 per million input tokens and $50 per million output tokens, less than half the price of the earlier Mythos Preview but still the most expensive of major AI models. For subscription users, Fable 5 is included in Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22. On June 23, Anthropic will remove it from those plans, requiring usage credits, with a plan to restore it as a standard subscription feature later.

Broader context

The release follows Anthropic's recent plea for global AI labs to establish a coordinated brake on frontier AI development, warning that systems may soon achieve recursive self-improvement without human intervention. Co-founder Jack Clark told the BBC that the industry has "a gas pedal, but it doesn't have a brake pedal." The launch also comes as Anthropic prepares to enter public markets, with its private valuation nearing $1 trillion.

You want the option to be able to take your foot off the gas and put your foot on the brake. Right now, it's like the AI industry has a gas pedal, but it doesn't have a brake pedal.