Two months after Anthropic released Mythos to a limited number of users, citing concerns about the intelligence model’s potential to wreak havoc in the wrong hands, the company said it was ready to release an equally powerful model to the public.
Anthropic on Tuesday announced the Claude Fable 5, a Mythos-class model that will be available to its business customers and paid subscribers. The company said the broader release is possible because of new safeguards that block responses in certain high-risk areas, including cybersecurity and biology.
“For us, it’s really close to what we call a ‘race to the top,’ to be able to provide this technology in a useful way, and at the same time we provide the right security measures so that they do equal benefits rather than harm,” Dianne Penn, Anthropic’s head of product management in research, told CNBC in an interview.
Anthropic attracted Wall Street and government officials in April with the unveiling of Mythos, which is highly effective at identifying security flaws within software. The company said it does not plan to make the model generally available, and has limited the release to a select group of companies as part of a cybersecurity program called Project Glasswing.
But with the launch of the Claude Fable 5, Anthropic is honoring its “ultimate goal” of bringing Mythos-class models to scale. It is also taking part in growing momentum and investor interest in its technology ahead of a potential IPO, which is expected to happen as soon as this year.
Anthropic said that Claude Fable 5 shows “amazing performance” in all software and information engineering tasks. In some benchmarks, it scored more than 10% higher than the Claude Opus 4.8, another model the company announced late last month, according to the blog.
Claude Fable 5 represents a “huge leap” in capability, which is why Anthropic had to implement additional safeguards to prevent misuse, Penn said. If the user asks a high-risk question, such as how to make ricin, a poison, for example, the model will block its answer and return to Claude Opus 4.8 to deliver a safe answer.
“What we wanted to do was to be very intentional about creating new types of categories and new types of security monitoring when launching this program,” said Penn.
Anthropic also introduced an updated model of the Mythos on Tuesday called the Claude Mythos 5. It’s the same basic model as the Claude Fable 5, but with raised protections in some areas, according to the blog.
Claude Fable 5 comes just days after Anthropic said it has filed a confidential IPO prospectus with the Securities and Exchange Commission, setting off a potentially historic selloff of shares after the company experienced a period of strong growth this year.
Anthropic said in May that revenue rose to $47 billion, up from about $10 billion in annual revenue last year. It recently closed a funding round worth $965 billion, surpassing its biggest competitor, OpenAI, which was valued at $852 billion in late March.
OpenAI is also gearing up for a major IPO, announcing on Monday that it has filed its prospectus privately with regulators.
Elon Musk SpaceXcombined with his AI startup xAI earlier this year, is set to make its public market debut on Friday.
Intense competition means that Anthropic CEO Dario Amodei will be under pressure to justify the startup’s valuation to investors, and Claude Fable 5 could be the company’s new significant money maker. The model costs $10 per million input tokens and $50 per million output tokens, which is twice as expensive as Claude Opus 4.8.
Penn said the prices are “too high” for customers, but that they don’t just want lower costs. He said users want high accuracy and high value for money, and that early customers of Claude Fable 5 have seen improvements in the use of each function.
“You get a higher ROI by having more intelligent models,” Penn said.
WATCH: Anthropic early investor Midha: There’s a whole universe of new opportunities on the AI frontier
