Are Claude models broken with the Fable 5 update?
Users report that Claude models are suddenly failing to work with tools (search, fetch, read, write) following the return of Fable 5 update.
Background
- Claude is a family of large language models (LLMs) made by Anthropic, a major US AI startup. Developers and power users can give Claude access to external "tools" (e.g., web search, file reading/writing, API calls) to let it perform real-world actions, not just chat.
- "Fable 5" is a new, obscure benchmarking framework (or "evals" suite) recently deployed on the platform. It seems to stress-test tool-calling under more complex, multi-step scenarios.
- Many users on Hacker News are reporting that since Fable 5 was rolled out, Claude models (especially the tool-calling variants) have become unreliable—failing on basic tool invocations that worked before. The implication is that the Fable 5 update may have accidentally broken the tool-use pipeline or changed the model's behavior in unintended ways.
- This matters because Claude is used by thousands of developers for automated coding, browsing, and data-processing tasks. A sudden regression in tool reliability undermines trust and disrupts workflows.