Thariq, posting on X, clarified that Claude Fable 5’s updated classifiers will still route a “small fraction” of routine coding and debugging requests back to Opus 4.8 when access is restored globally tomorrow.
In the follow-up posts, Thariq states that the team is continuing to refine the safeguards to better distinguish “genuine misuse” from legitimate requests and to reduce false positives. When asked whether any fallback would be visible to users, Thariq replied that it “will be transparent.”
He also indicated that if a request is routed to Opus, it will be “billed at opus rate, with prompt cache miss paid for.” In response to a question about why some ordinary prompts were being downgraded, Thariq wrote that these are “false positives that we don’t want to flag but get caught with the classifier so it’s not quite deterministic.”
The clarification follows a post from Anthropic saying that Fable 5 will return globally after “a series of productive conversations with the US government,” with the company adding that it is redeploying the model with new classifiers aimed at blocking more cybersecurity tasks. That announcement also introduced the wording around some routine coding and debugging work falling back to Opus 4.8, which drew immediate questions from users about availability, billing, and how broad the restrictions might be.
Source: X post by Thariq
