Key facts
- Claude Fable 5's performance has not degraded.
- User and benchmark concerns about performance have been raised.
- A new safety classifier has been implemented.
- The safety classifier reroutes coding and debugging tasks.
- Rerouted tasks are sent to a different model.
- This rerouting has led to perceived quality drops in coding and debugging.
- Anthropic states the core model's capabilities are unaffected.
- The rerouting is described as a temporary measure.
Anthropic has stated that the performance of its Claude Fable 5 model has not degraded, despite recent user reports and benchmark concerns suggesting otherwise. The company attributes the perceived drops in quality to a new safety classifier that has been implemented. This classifier is rerouting a significant number of coding and debugging tasks to a different, unspecified model. Anthropic clarifies that the core capabilities of Claude Fable 5 remain intact and that the rerouting is a temporary measure. The company is working to refine the safety classifier to better handle these types of tasks without impacting perceived performance. Users experiencing issues with coding or debugging tasks are advised that this rerouting is the cause and that Anthropic is actively working on a solution. The specific details of the alternative model used for rerouting and the exact nature of the safety concerns have not been disclosed. Anthropic assures users that the integrity and performance of Claude Fable 5 are not compromised at a fundamental level, and the current situation is a result of a specific feature adjustment aimed at enhancing safety protocols.