Anthropic’s Mythos goes public—on a short leash

TL;DR: On Tuesday, Anthropic released Claude Fable 5, a version of its vulnerability-hunting Mythos model it says is safe for public use. According to the company, Fable 5 is state-of-the-art on nearly every benchmark it tested. But it’s still a leashed version of Mythos: Ask anything in areas like cybersecurity and biology, and you’ll be rerouted to the older Claude Opus 4.8 model.

What happened: The model is now free on Pro, Max, Team, and seat-based Enterprise plans through June 22, after which it will require usage credits. Fable 5 is the first Mythos-class model Anthropic has made generally available, with the highest scores across some coding and finance benchmarks, according to the company.

Initial tests: Early reviews indicate that Fable 5 is a monster that’s exhausting to feed. After spending five-plus hours testing it and running up a $110 token bill in a single day (that was fortunately covered as part of his subscription), developer Simon Willison called it “something of a beast. It’s slow, expensive and has been quite happily churning through everything I’ve thrown at it so far.”

Wharton professor Ethan Mollick, who had early access, was impressed to see it spin up its own swarm of agents to research and check one another’s work, but griped that the guardrails sending his queries to Opus 4.8 tripped “way too often.” (Anthropic conceded the safeguards are tuned conservatively and will “sometimes catch harmless requests.”)

Timely flex: All eyes are on how Anthropic justifies its $965 billion valuation ahead of its planned IPO. Fable 5 isn't cheap—it costs $10 per million input tokens and $50 per million output tokens (double that of Opus 4.8), but that price tag could make it a tidy new earner for Anthropic.

“For us, it’s really around what we call ‘race to the top,’ being able to provide this technology in a valuable fashion,” Dianne Penn, Anthropic’s head of product management for research, told CNBC. “And at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm,”

Mythos 5, which has the same underlying model as Fable 5 with some safeguards lifted, launches first for a small group of cyberdefenders and infrastructure providers through Project Glasswing, Anthropic said, before a wider rollout.

Bottom line: Anthropic spent two months calling Mythos too dangerous to ship. Now, a guardrailed version is here, and the company is betting its most-feared model can go mainstream without blowing up—and make it a lot of money in the process. —LC