Back in April, Anthropic presented its “Methos” model to the world.. Mythos Preview, allegedly, is. A powerful model that can find security flaws in all types of software. In the wrong hands, bad actors can abuse the model to find vulnerabilities in the programs, services, and sites that most of us rely on for modern digital life. In fact, Mythos may open the biggest hacking opportunity in history. what a pitch
Thus, Anthropic pulled the brakes on Mythos. Although it maintained that it would eventually release the model to the public, it needed it first. Test it with a limited pool of trusted testers.calling it “Project Glass Wing”. To begin with, this meant opening up the model to the US and other governments. While Mythos is not yet available for the likes of you or me, Anthropic. is Releasing a new model that promises many of the capabilities of Mythos, without the cybersecurity risks.
What are Anthropic’s Fable 5 and Mythos 5?
On Tuesday, Anthropic announced its latest model, Claude Fable 5which it calls a “Mythos-class model” that is “safe for general use.” The company says the Fable 5 is better and more capable than its other public models. Anthropic claims that the Fable 5 scores at the top of most benchmarks, including software engineering, cognitive tasks, vision tasks, and research. “The longer and more complex the task, the larger Fable 5 will have an edge over our other models,” the company says. There’s also Mythos 5, which seems to be Fable 5 without some limitations, but isn’t available to the general public.
According to Anthropic’s benchmarking, Fable 5 and Mythos 5 equally outperformed Mythos Preview, Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, in the following categories: agentive coding, cognitive tasks, spatial reasoning, tool use, legal, multidisciplinary tool, toolout, and health. The Mythos Preview wins in computer use and multidimensional reasoning (with tools), but it’s a clean sweep of all other models.
Credit: Anthropic
Anthropic says that Fable 5 was able to complete a coding project that would have taken a team over two months to complete in just one day. It can reconstruct the source code of a web app from screenshots only. It can beat. Pokemon Fire Red with “minimal, vision-only usage”, while other cloud models struggled to play at all. It was playable. Kill the spire And Opus 4.8 Mythos 5 triples its research capabilities to the final process, with improved statistics in drug design, as well as the ability to generate new hypotheses regarding questions in molecular biology, and new research in genomics.
How does Anthropic secure Fable 5?
Here’s the big question: if Fable 5 is Mythos-class, how can you make sure it’s safe to release to the general public? Couldn’t a bad actor take advantage of Fable 5’s capabilities and force it to discover and reveal security vulnerabilities?
Anthropic says it has figured it out. While Fable 5 may be Mythos-level in many ways, the company says its Project Glasswing testing has produced a model with adequate security measures for public release. Myth 5 explores “classifiers,” or highly sensitive topics, that he knows he shouldn’t answer. This means: When Fable 5 receives a request that it thinks has to do with cybersecurity, biology, chemistry, or distillation, it doesn’t answer the question itself. Instead, it forwards the query to Opus 4.8, Anthropic’s “next-most-capable” model. The model must still be powerful enough to provide accurate answers, but not able to provide malicious users with the necessary tools to exploit others.
What do you think so far?
Anthropic says its new guardrails are cautious and conservative, and can be hit to extremes. Benign requests can accidentally trip the Fable 5’s security alarm, but this supposedly only happens about 5% of the time. As such, Anthropic says the Fable 5 is able to handle requests on its own about 95% of the time. In addition, the company found that after the bug bounty program, no white-hat hacker could find a universal jailbreak (or an exploit to bypass security protocols) after 1,000 hours of testing. While one organization has made progress in finding a jailbreak, Anthropic says it believes its protocols make it impractical for hackers to discover a jailbreak before the company does.
Why Skip Biology and Chemistry Applications? Anthropic says Mythos is also great at helping gene therapy research and development, which can be beneficial for scientists, but a big risk in the wrong hands. Also, Anthropic knows that there are actors out there trying to “simplify” the capabilities of cloud models to train their models to do whatever they want. Thus, any of these applications are booted on a low-performance model.
Anthropic is also changing its data retention policy for Fable 5 and Mythos 5. With these models, the company will keep your data for 30 days — not for training, but to help protect against future cyberattacks and jailbreaks. Both Fable 5 and Mythos 5 cost the same: $10 per million input tokens, and $50 per million output tokens, which Anthropic says is less than half the cost of the Mythos Preview.