Claude Fable 5 and Claude Mythos 5 are two new fifth-generation AI models released by Anthropic on June 9, 2026. Both share the same underlying model, which Anthropic places in a new “Mythos-class” tier that sits above its Opus models in raw capability. Fable 5 is the version built for the general public, shipping with conservative safety guardrails, while Mythos 5 is the same model with some of those restrictions lifted, reserved for a small group of trusted cyberdefenders and researchers. Together they represent Anthropic’s attempt to put frontier-level performance into everyday hands while keeping its most dangerous capabilities behind a gate.
In this article we’ll discuss what makes these models different from earlier Claude releases, how Fable 5 performs across coding, knowledge work, and science, why Anthropic built an unusual set of safeguards to release it, and what the rollout means for the people who’ll actually use it. We’ll also look at the pricing, the staged availability plan, and the broader strategy behind splitting one model into two products with two names.
TL;DR Snapshot
Anthropic has launched its first publicly available “Mythos-class” model, a tier the company says exceeds anything it has shipped to the general public before. The headline release, Claude Fable 5, posts state-of-the-art scores across software engineering, vision, finance, and scientific reasoning benchmakrs, and it can work autonomously on long, complex tasks for longer than any previous Claude offering. Its sibling, Mythos 5, is the same model with cyber and biology guardrails removed, and is only available to vetted partners because its capabilities (especially in cybersecurity) could cause real damage in the wrong hands.
Key takeaways include…
- Fable 5 and Mythos 5 are the same underlying model. The only difference is the safeguards, which is also why they have different names.
- Fable 5 leads or ties on nearly every benchmark Anthropic tested, and its lead grows the longer and more complex a task gets.
- When Fable 5 detects a sensitive request (cybersecurity, biology, chemistry, or model distillation), it quietly hands the response off to Claude Opus 4.8 instead of refusing to process it outright.
Who should read this: Developers, founders, AI enthusiasts, and anyone tracking how frontier AI gets released safely.
What “Mythos-Class” Actually Means
The most important thing to understand is that “Mythos-class” is a capability tier, not a single product. According to Anthropic’s announcement, Mythos-class models sit above the company’s Opus class in capability, and the first of them, Claude Mythos Preview, was released back in April to a limited group of cyber defenders and infrastructure providers.
Fable and Mythos are two new entries into that same tier. As Anthropic explains in their launch post, the name “Fable” comes from the Latin fabula, meaning “that which is told,” which is closely related to the Greek mythos. Fable 5 is the safety-tuned version that anyone can use, and Mythos 5 is the restricted version reserved for trusted partners. The Decoder reports that both models share the same base, with Fable shipping conservative guardrails for general use and Mythos dropping those restrictions in areas like cybersecurity for a small group of partners.
How Fable 5 Performs
Anthropic claims that Fable 5 is state-of-the-art on nearly all tested benchmarks, and the numbers it published back that up, especially when it comes to coding. On SWE-Bench Pro, a test that measures solving real software engineering tasks pulled from public GitHub repositories, The Decoder reports that Fable 5 hits 80.3 percent, compared with 69.2 percent for Claude Opus 4.8, 58.6 percent for GPT 5.5, and 54.2 percent for Gemini 3.1 Pro. On Cognition’s FrontierCode benchmark, which tests demanding coding tasks under production standards, the same report notes Fable 5 scores 29.3 percent against 13.4 percent for Opus 4.8 and just 5.7 percent for GPT 5.5.
The real-world examples are just as striking. In early testing, Anthropic says Stripe used Fable 5 to compress months worth of engineering into days, performing a complex migration across a 50-million-line Ruby codebase with unprecedented speed. It’s estimated that it would have taken a whole team of human engineers more than two months to complete the task by hand.
It doesn’t just excel at coding though, Anthropic reports strong gains in knowledge work and vision too, including the ability to rebuild a web app’s source code from screenshots alone. Several early customers echoed that sentiment. Cursor CEO Michael Truell described the model as “the state of the art model on CursorBench,” adding that it opened up a class of long-horizon problems that earlier models couldn’t reach.
Why Anthropic Built Safeguards Into the Release
Here’s the catch that shapes the whole release. A model this capable is genuinely risky, so Anthropic didn’t ship Fable 5 with the brakes off. Instead it built a layer of classifiers, separate AI systems that watch for misuse, and when those classifiers flag a request, Fable 5 doesn’t refuse. It hands the response off to Claude Opus 4.8 instead, and the user gets told when that happens.
The classifiers cover four key areas: cybersecurity, biology, chemistry, and model distillation. The reasoning is that Mythos-class models are unusually good at finding and exploiting software vulnerabilities, and that same skill could make cyberattacks cheaper and easier to pull off. CNBC reported that the broad public release was only possible because of these new safeguards blocking responses in high-risk areas.
Anthropic tuned the safeguards cautiously, and by its own admission they’re stricter than ideal, so they’ll sometimes catch harmless requests. Even so, the company says they trigger on average in less than 5 percent of sessions, which means more than 95 percent of Fable sessions involve no fallback at all. To stress-test the system, Anthropic ran an external bug bounty that produced no universal jailbreaks across more than 1,000 hours of testing. There’s also a new data policy attached to the release, Anthropic will require 30-day retention for all traffic on Mythos-class models, which it says it won’t use for training and will use only for safety monitoring.
The Science Story Behind Mythos 5
While Fable 5 grabbed the public spotlight, the most promising results in the scientific sphere came from Mythos 5, the restricted-access sibling. According to Anthropic, their internal protein design experts accelerated parts of the drug design process by around ten times using Mythos 5, and the model produced strong drug-design candidates for 9 of the 14 protein targets it was tested against, working without human assistance on tasks normally done by a trained scientist.
Anthropic also reports that Mythos 5 is its first model to consistently generate novel, compelling scientific hypotheses, and in blind head-to-head comparisons their scientists preferred Mythos’s molecular biology hypotheses about 80 percent of the time. In one genomics project, Mythos worked largely autonomously for over a week, then trained a custom machine learning model that outperformed a recent model published in the journal Science, despite being 100 times smaller.
Model Pricing and Availability
Both models are priced identically at $10 per million input tokens and $50 per million output tokens, which Anthropic notes is less than half the price of Claude Mythos Preview. As the AWS team confirmed, Claude Fable 5 is also available on Amazon Bedrock and the Claude Platform on AWS from launch day.
The subscription rollout is more cautious, because Anthropic expects demand to be very high and hard to predict. From launch through June 22, Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans. On June 23, it gets removed from those plans, and continued use will require usage credits until capacity expands. Anthropic says it intends to restore Fable 5 as a standard part of subscription plans as quickly as it can.
Mythos 5, meanwhile, stays locked down. Anthropic explains that it’s restricted to Project Glasswing partners only at the moment, and soon to a small set of biology researchers. Nobody else will be able to use it until a broader trusted access program opens up.
Frequently Asked Questions
It’s a capability tier that Anthropic created to sit above its Opus-class models. They are its most powerful models, and Claude Mythos Preview (released in April) was the first one, followed now by Fable 5 and Mythos 5.
They’re the same underlying model. Fable 5 ships with conservative safety guardrails in place and is available to the public, while Mythos 5 has some of those guardrails lifted and is restricted to trusted partners. The safeguards are the only thing that separates them.
It’s Anthropic’s program for releasing Mythos-class capabilities to a limited group of trusted cyberdefenders and infrastructure providers. It’s a collaboration that has included companies such as Amazon Web Services, Apple, Google, Cisco, Microsoft, and JPMorgan Chase, who got early access to find and fix vulnerabilities in their own software.
Instead of refusing to process the users request, Fable 5 hands it off to Claude Opus 4.8, and the user is told when this happens. Anthropic says this affects less than 5 percent of sessions on average.
A jailbreak is a prompt or technique that lets a user interact with a model as if its safeguards weren’t there. A “universal” jailbreak is one that works broadly rather than in a single narrow situation. Anthropic says its external bug bounty found no universal jailbreaks in over 1,000 hours of testing.
Distillation is the practice of extracting a model’s capabilities to train a competing model. Anthropic flags distillation attempts as one of the categories its classifiers watch for, because distilling Fable 5 could spread near-frontier capabilities without the matching safeguards.
AI models charge based on tokens, which are small chunks of text (roughly a few characters each). Fable 5 and Mythos 5 cost $10 per million input tokens (the text you send) and $50 per million output tokens (the text the model generates back).
Other Enterprise AI Articles You May Be Interested In
Apple’s New Siri AI: Everything You Need to Know About the Biggest Siri Overhaul in Over a Decade
Why Corning’s Amazon Deal Makes It the Most Important AI Infrastructure Stock of 2026
Broadcom Stock Plunges on Soft AI Semiconductor Guidance: Full Breakdown
Publishers Can Now Opt Out of Google AI Overviews: Inside the UK’s Historic Ruling
Claude Opus 4.8: Everything You Need to Know About Anthropic’s New AI Model
