Claude Fable Relaunch Disappoints Users With Stricter Safety Guardrails and Usage Restrictions
Key Takeaways
- ▸Claude Fable 5 is now available to all subscribers but is capped at 50% of weekly usage limits before transitioning to usage-based credits on July 7
- ▸Users report significant performance degradation with frequent false-positive safety guardrails triggering fallbacks to Opus 4.8
- ▸The model is overly restrictive on legitimate coding tasks, particularly those involving security-adjacent language or systems-level programming
Summary
Anthropic has relaunched Claude Fable 5, its most powerful model, following the Department of Commerce's lifting of a previous ban. The model is now available to all users across Max, Pro, and Team subscription tiers, but with significant restrictions: users can only access Fable for up to 50% of their weekly usage limits, and the model will transition to a pay-as-you-go credit system starting July 7.
However, the relaunch has disappointed users who report substantial performance degradation compared to the original release. On Reddit and other developer forums, users describe the model as "nerfed," frequently triggering overly aggressive safety guardrails that cause the model to fall back to Opus 4.8 on legitimate tasks. Developers report that the model refuses tasks involving systems-level coding, C/C++, Rust, security-related language, and memory references—even when the actual task poses no safety risk.
Anthropologic has indicated that the model itself has not been technically nerfed, but rather that it now operates with conservative safety margins. Nevertheless, Anthropic has not yet publicly addressed reports of false positives, though the company is likely aware of the performance complaints.
- Anthropic says the model itself is unchanged but safety margins have been conservatively increased, though the company has not yet publicly acknowledged the false-positive complaints
