Two months after Anthropic rolled out Mythos to a limited number of users, citing concerns that its artificial intelligence model could cause damage in the wrong hands, the company has announced that it is ready to release an equally powerful model to the public.
Anthropic on Tuesday announced Claude Fable 5, a Mythos-class model available to enterprise customers and paid subscribers. The company said widespread release was possible thanks to new safeguards that block responses in certain high-risk areas, such as cybersecurity and biology.
“For us, it’s really about what we call the ‘race to the top,’ being able to deliver this technology in a way that’s valuable, but also providing the right safety guardrails that asymmetrically do more good than harm,” Diane Penn, head of research product management at Anthropic, told CNBC in an interview.
Anthropic introduced Mythos in April, which excels at identifying security flaws in software, captivating Wall Street and government officials. The company said it has no plans to make this model generally available and is limiting its rollout to a select group of companies as part of a cybersecurity initiative called “Project Glasswing.”
But with the release of Claude Fable 5, Anthropic is adhering to its stated “end goal” of deploying the Mythos class of models at scale. The company is also capitalizing on growing momentum and investor interest in its technology ahead of a major IPO planned for early this year.
Anthropic said Claude Fable 5 has demonstrated “exceptional performance” across software engineering and knowledge work tasks. In some benchmarks, it scored more than 10% higher than another model, the Claude Opus 4.8, which the company announced late last month, according to a blog post.
Claude Fable 5 represents a “huge leap forward” in functionality, which required Anthropic to implement additional guardrails to prevent exploitation, Penn said. For example, if a user asks a risky question, such as how to make the toxin ricin, the model blocks the response and falls back to Claude Opus 4.8 to provide a safe answer.
“What we wanted to do was be very intentional about building a new type of classifier and a new type of safety guardrail for this launch,” Penn said.
Anthropic also announced its latest Mythos model on Tuesday, called the Claude Mythos 5. According to the blog post, it’s the same basic model as Claude Fable 5, but with protective measures removed in some areas.
Claude Fabre 5 was announced just days after Anthropic announced it had secretly filed an IPO prospectus with the Securities and Exchange Commission, setting up a potentially historic stock sale after experiencing a period of explosive growth this year.
Anthropic announced in May that its revenue run rate had ballooned to $47 billion, up from about $10 billion in annual revenue last year. The company recently completed a funding round at a valuation of $965 billion, surpassing its biggest rival, OpenAI, which was valued at $852 billion in late March.
OpenAI is also gearing up for a major IPO, announcing on Monday that it had confidentially filed a prospectus with regulators.
Elon Musk’s space xmerged with his AI startup xAI earlier this year and is planning a record-breaking public market debut on Friday.
Fierce competition means Anthropic CEO Dario Amodei will be under pressure to justify the startup’s valuation to investors, and Claude Fable 5 could be a valuable new cash cow for the company. This model costs $10 per million input tokens and $50 per million output tokens, which is twice the cost of Claude Opus 4.8.
Penn said pricing is a “top priority” for customers, but they aren’t just looking for cost savings. He said users are demanding more precision and greater returns for the money they spend, and early customers of Claude Fable 5 have noticed improved spend per task.
“Having a more intelligent model just gives you a higher ROI,” Penn says.
Featured: Early Human Investor Mida: There’s a whole ecosystem of new opportunities on the AI frontier
