ConTEB
A public benchmark designed specifically for evaluating contextual embedding capabilities. It constructs scenarios requiring cross-chunk understanding for correct retrieval, measuring an embedding model's ability to handle cross-chunk dependencies within a document. It is distinct from the more general-purpose MTEB benchmark.
Timeline (last 90 days)
pplx-embed-v2-context-9b-preview has achieved state-of-the-art (SOTA) results on two benchmarks: ConTEB and context-bench.
ConTEB is a publicly available evaluation set specifically designed to assess contextual embedding capabilities, distinct from the more general MTEB.
pplx-embed-v2-context-9b-preview achieved best-in-class performance on both the ConTEB and turbopuffer context-bench benchmarks (official data).
All Facts (3)
pplx-embed-v2-context-9b-preview has achieved state-of-the-art (SOTA) results on two benchmarks: ConTEB and context-bench.
50%UnverifiedConTEB is a publicly available evaluation set specifically designed to assess contextual embedding capabilities, distinct from the more general MTEB.
50%Unverifiedpplx-embed-v2-context-9b-preview achieved best-in-class performance on both the ConTEB and turbopuffer context-bench benchmarks (official data).
50%