Anthropic launched Claude Fable 5, its most capable consumer-facing AI model to date, but with a catch: some sensitive cybersecurity requests are automatically downgraded to a less powerful model.

The company said Fable 5 excels at coding, research and other knowledge-intensive tasks, particularly over long and complex workloads.

To curb potential abuse, Anthropic built in guardrails that redirect certain high-risk queries to Claude Opus 4.8. The filters activate in fewer than 5% of sessions on average, according to the company.

Mythos 5 uses the same underlying model as Fable 5 but drops some of the restrictions—for a select group of cyber defenders, critical infrastructure operators and other trusted partners.

The company said Mythos 5 will first be delivered through Project Glasswing in partnership with the U.S. government as an upgrade from Claude Mythos Preview, with broader access planned through a trusted program.

Anthropic set pricing for both Fable 5 and Mythos 5 at $10 per million input tokens and $50 per million output tokens. The company said that rate is less than half the price of Claude Mythos Preview.

Pokémon By Day, Protein Design By Night

The models can sustain longer autonomous work than prior Claude versions and can stay on track across very long contexts while using notes to refine outputs. The company described tests in games such as "Pokémon FireRed" and "Slay the Spire" as demonstrations of those capabilities.

On the life sciences side, internal protein design specialists used Mythos 5 to speed up parts of drug design work by roughly tenfold, and that several targets produced promising candidates now under review. 

The company also noted that Mythos 5 has generated original hypotheses in molecular biology and completed a week-plus genomics project that produced a custom model it said exceeded a recently published approach despite being far smaller.

An automated alignment assessment found low levels of problematic behavior for Mythos 5, similar to Opus 4.8, and it expects comparable results for Fable 5 given the shared core model. 

Anthropic previously shared an update on a security-focused collaboration, Project Glasswing. Its artificial intelligence-assisted security testing effort had already uncovered “more than 10,000 high-or critical-severity vulnerabilities” across widely used software systems.

Project Glasswing previously included roughly 50 partner organizations and later expanded access to nearly 150 organizations.

Anthropic IPO Plans

Last week, Anthropic confidentially filed a draft Form S-1 with the U.S. Securities and Exchange Commission, giving the artificial intelligence company the option to go public after the SEC completes its review.

Anthropic said it has not yet determined how many shares it would sell or what price it would seek in an offering.

Last month, Anthropic overtook OpenAI as the world's most valuable startup after raising $65 billion in Series H, valuing the company at $965 billion. Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital led the funding round.

If Anthropic goes public at a $1 trillion valuation, it would instantly rank among the most valuable companies globally and could become the second- or third-largest IPO in history, trailing SpaceX and Saudi Aramco.

Photo: Shutterstock