Anthropic has officially released Claude Fable 5, a public version of its previously restricted "Mythos Preview" model, which the company first unveiled in April. According to Tom's Hardware, Anthropic bills the new release as "safe for general use" and claims it is "state-of-the-art on nearly all tested benchmarks."
But real-world reception has been bumpier. Security firm Endor Labs published an analysis arguing that Fable 5 delivers "mid-tier results on coding tasks" — a verdict pointedly packaged under the headline "Mythos-grade hype." The piece gained traction on Hacker News, where it earned 168 points and drew nearly 70 comments from developers weighing in on the gap between Anthropic's claims and their own experience.
A separate grievance is also mounting. According to Fast Company, developers say Fable 5 "plays it too safe on safety" — meaning its built-in guardrails are so conservative that the model becomes difficult to use for practical software work. Overly cautious refusals are a known frustration in the developer community, where productivity depends on a model that can engage with complex, sometimes ambiguous requests without flinching.
The pushback puts Anthropic in a familiar bind. The company has staked much of its identity on responsible AI development, but a model that aces controlled benchmarks while irritating working developers raises questions about what those benchmarks are actually measuring.
As AI coding assistants become embedded in everyday software development, the distance between lab performance and practical usefulness is shaping up to be the central fight among frontier AI labs.