Linkup Research Releases SPARSEUP: A 149M-Parameter Open-Source Sparse Embedding Model

Linkup Research has released SPARSEUP, an open-source sparse embedding model built on a 149M-parameter ModernBERT backbone.

Benchmark Performance

SPARSEUP scores 56.4 nDCG@10 on BEIR-13, which Linkup calls the best result it knows of for a public sparse encoder under 150M parameters.

Model Design

The model uses the following techniques to keep its vectors sparse:

• A logit shift
• Top-12 expansion per token
• Case folding

Retrieval Speed and Recall

With the Seismic index, SPARSEUP reaches over 97% recall in about 380 microseconds per query.

Licensing

The model ships under Apache 2.0.