Anthropic releases Claude Fable 5, its most capable public model yet

Anthropic spent the spring warning that frontier AI is getting genuinely dangerous. On June 9 it shipped its answer to that tension: Claude Fable 5, a model the company describes as Mythos-class capability made safe enough for everyone.

The headline claims are not shy. State of the art on nearly every benchmark Anthropic tested, across software engineering, knowledge work, vision and scientific research. The examples in the announcement include a 50-million-line codebase migration completed in a single day and protein design work accelerated roughly tenfold. The context window is one million tokens.

The pricing is the strategic move. At $10 per million input tokens and $50 per million output, Fable 5 lands at less than half the cost of Claude Mythos Preview, and it is bundled into Pro, Max, Team and seat-based Enterprise plans at no extra charge from June 9 through June 22. Anthropic wants everyone to try this thing, immediately.

The safety architecture is the other story. New classifiers watch for high-risk territory, cybersecurity exploitation, biology and chemistry, attempts to distill the model, and quietly route those sessions to Claude Opus 4.8 instead. Anthropic says this triggers in under 5 percent of sessions. TechCrunch reports the model survived an external bug bounty with more than a thousand hours of jailbreak attempts and no universal exploit found.

There is also a sibling. Claude Mythos 5, the same underlying model with certain safeguards lifted, goes only to vetted cyberdefenders and infrastructure operators through a program called Project Glasswing, run with the US government.

For creative teams, the practical takeaway is simple: the strongest generally available model on the market just got cheaper than its preview tier, and for two weeks it costs nothing extra to find out what it does to your workflow.