Key Info

A production benchmark from Rox compared LLM-based reranking against Jev classification for retrieving sales data (transcripts, emails, CRM notes, news, documents), reporting Jev is 20x faster, 10x cheaper, and 12% more accurate than GPT-5 Mini.

Highlights

  • Benchmark scenario: every Rox agent query pulls from transcripts, emails, CRM notes, news, and documents
  • Two retrieval approaches compared: LLM-based reranking vs. Jev classification
  • Jev reported 20x faster than GPT-5 Mini baseline
  • Jev reported 10x cheaper than GPT-5 Mini baseline
  • Jev reported 12% higher accuracy than GPT-5 Mini baseline