claude-code - 💡(How to fix) Fix Auto mode: classifier that picks model (Haiku/Sonnet/Opus) and thinking effort per task [1 participants]

TL;DR

Implement an "auto" mode that uses a classifier to automatically select the model and thinking level based on task characteristics.

Guidance

• Develop a classifier that can assess task complexity, breadth, novelty, and safety stakes to determine the optimal model and thinking level combination.
• Integrate the classifier into the system, using a lightweight model like Haiku to minimize latency and cost.
• Display the chosen configuration in the UI to allow users to learn from and override the auto mode when necessary.
• Consider implementing a feedback mechanism to improve the classifier's accuracy over time.

Example

No code snippet is provided as the issue does not contain specific technical details.

Notes

The proposed solution relies on the development of an effective classifier, which may require significant testing and refinement to ensure accurate model selection.

Recommendation

Apply workaround: Implement the proposed "auto" mode with a classifier to simplify model selection and reduce user friction. This approach addresses the core issue of model selection complexity and provides a sensible default for new users.