LiteLLM Launches FUSE V2 AutoRouter: Predictive Model Selection Cuts Cost on SWE-bench

LiteLLM (YC W23) ·

Key Info

LiteLLM has announced FUSE V2 for its AutoRouter, a model-routing layer that predicts which model is most likely to succeed on a given task. In SWE-bench testing, FUSE V2 solved 25 out of 25 tasks while lowering the cost per solved task.

Highlights

  • FUSE V2 predicts the best model for each task to improve routing efficiency and cost.
  • Early benchmark results show 25/25 tasks solved on SWE-bench.
  • The approach focuses on both task success and reducing the cost per solved task.
Loading...