16 November 2026

LlamaIndex's ParseBench-2 benchmark faces bug and self-favoritism accusations

First reported

Latent Space ran this on .

  • LlamaIndex released ParseBench-2, a benchmark for document-parsing software, and rivals Extend and Datalab called it self-favoring.
  • Extend and Datalab said a bug in an adapter changed scores sharply, which they argue undermines the results.

How it was covered