LLM reasoning, automated: tokens drop 69.5%

Researchers from Meta, Google, and universities introduced AutoTTS, a framework that automatically discovers test-time scaling strategies for LLMs.

Instead of manually designing branching, pruning, or stopping rules, AutoTTS uses an explorer model to iteratively refine controllers in a defined control space.

In experiments, the system reduced token usage by up to 69.5% while maintaining performance.

The approach could lower inference costs and reduce the need for manual heuristic tuning in reasoning models.