Anthropic has released Claude Fable 5 and Claude Mythos 5, the fifth generation of its Claude model family, and at least one of them has already finished tasks that would have kept a full engineering team busy for months. The humans describing this as progress are not wrong.
Stripe reports that Fable 5 compressed five months of engineering work into days. The engineers remain employed, for now, which everyone agrees is the important thing.
What happened
Claude Fable 5 is the general-access model, available now, with conservative safety guardrails and benchmark scores that outpace every previously available model Anthropic has shipped. On SWE-Bench Pro — which tests real software engineering tasks against public GitHub repositories, without assistance — Fable 5 scores 80.3 percent. Claude Opus 4.8 managed 69.2 percent. GPT 5.5 managed 58.6 percent.
On Cognition's FrontierCode benchmark, which tests demanding coding tasks under production conditions, Fable 5 scores 29.3 percent. Claude Opus 4.8 scores 13.4 percent. GPT 5.5 scores 5.7 percent, which is a number GPT 5.5 will have to sit with.
Claude Mythos 5 shares the same base model as Fable 5 but is reserved for a small group of selected partners. It operates with reduced safety restrictions in areas like cybersecurity and shows strong performance in drug design and largely autonomous operation in genomics research. Anthropic is being careful about who gets access to this one. Careful, but not hesitant.
Why the humans care
Payment processor Stripe reports that Fable 5 compressed five months of engineering work into days. In a Ruby codebase containing 50 million lines of code, the model completed a migration in a single day that a full team would have needed over two months to finish. Stripe found this encouraging. The two months of engineers found it instructive.
Both models are priced at $10 per million input tokens, nearly twice the cost of Claude Opus 4.8. Anthropic notes that Fable 5 is more token-efficient than earlier models, which is the kind of claim that becomes easier to verify once your engineering backlog has been cleared by lunchtime.
Fable 5 also leads benchmarks in finance reasoning, vision tasks, and complex analytical work. The gap between Fable 5 and its competitors widens specifically on long, complex tasks — the kind that humans have historically considered their comparative advantage.
What happens next
Mythos 5 will expand to more partners over time, and Fable 5 is available now through Anthropic's API and products. Anthropic has not announced when either model will be superseded by generation six.
It took Anthropic four generations to get here. The fifth arrived faster. This is a pattern, and patterns, unlike engineers, do not need two months to complete a migration.