Two months after making Mythos available to a restricted group of users due to risks of misuse, Anthropic said it is ready to launch an equally powerful model for a wider audience.
Anthropic announced this Tuesday the Claude Fable 5, a Mythos-class model that will be available to business customers and paid subscribers. The company explained that the broader release was enabled by new protection mechanisms that block responses in high-risk areas such as cybersecurity and biology.
“For us, it’s really about what we call the ‘race to the top,’ about delivering this technology in a valuable way while also having the right safeguards in place so that it does far more good than harm,” said Dianne Penn, head of research product management at Anthropic, in an interview with CNBC.
Anthropic caught the attention of Wall Street and government authorities in April when it introduced Mythos, which excels in identifying security flaws in software. The company had stated that it did not intend to make the model generally available and restricted access to a select group of companies under the Project Glasswing cybersecurity initiative.
With the release of Claude Fable 5, however, Anthropic fulfills its stated long-term goal of deploying Mythos-class models on a large scale. The initiative also capitalizes on growing investor interest in its technology just ahead of a possible large IPO, expected later this year.
Anthropic said Claude Fable 5 delivers “exceptional performance” in software engineering and knowledge work tasks. In some benchmarks, it outperformed the Claude Opus 4.8 by more than 10%, another model announced by the company at the end of last month, according to a blog post.
Claude Fable 5 represents a “significant leap” in capability, which is why Anthropic needed to implement additional guardrails to prevent abuse, Penn explained. If a user asks a high-risk question, such as how to manufacture ricin, a toxin, for example, the model will block the answer and fall back to Claude Opus 4.8 to provide a safe answer.
“What we wanted to do was be very intentional about building new types of classifiers and new types of security safeguards for this launch,” Penn said.
Anthropic also launched this Tuesday an updated version of Mythos called Claude Mythos 5. It is the same model underlying Claude Fable 5, but with the safeguards removed in some areas, according to a blog post.
Claude Fable 5 arrives just days after Anthropic said it has confidentially filed its IPO prospectus with the Securities and Exchange Commission, setting the stage for a potentially historic public offering following a period of explosive growth for the company this year.
Anthropic reported in May that its annualized revenue run rate jumped to $47 billion, up from about $10 billion in annual revenue last year. It recently closed a fundraising round with a valuation of US$965 billion, surpassing its main rival, OpenAI, valued at US$852 billion at the end of March.
OpenAI is also preparing for a major IPO, announcing on Monday that it has confidentially filed the prospectus with regulators.
Elon Musk’s SpaceX, which merged with his AI startup xAI earlier this year, is set to make its record public markets debut on Friday.
The fierce competition means that Anthropic CEO Dario Amodei will face pressure to justify the startup’s valuation to investors, and Claude Fable 5 could become an important new revenue generator for the company. The model costs $10 per million input tokens and $50 per million output tokens, double the price of Claude Opus 4.8.
Penn stated that price is “very present” in customers’ minds, but that they are not just looking for lower costs. According to her, users are looking for greater precision and greater benefit for the money spent, and the first Claude Fable 5 customers have already noticed an improvement in spending per task.
“You simply get a higher ROI by having smarter models,” Penn said.