Advanced retrieval models
Put the right answer first — with Rerank
Rerank is a powerful model that gives a semantic boost to search quality. Drop it in after any retrieval step to reorder candidates by true relevance — so the best result lands at the top, every time.
Relevance lift by query type
Powering search for the teams building what people look for next
Sharper relevance, every query
Rerank reads each query and candidate together, then reorders the list by genuine meaning — turning a rough shortlist into a precise answer.
5/12, 9:15 AM
Rerank, reorder these results for "how do I reset my password".
Harriet 9:16 AM
@Rerank which of these passages best answers the billing question?
Rerank APP 9:16 AM
Re-scored every candidate — here is what changed:
- Top result moved from rank 9 to rank 1 after re-scoring
- Two near-duplicate passages collapsed below the better match
- An off-topic candidate dropped out of the top ten entirely
Search accuracy lift
0.0x
Find the right answer faster
Teams report more than double the top-result accuracy after adding Rerank to their pipeline.
Better answers, lower effort, near-zero overhead. A semantic boost that pays off on every single query.
Want the full picture on Rerank quality gains?
Download the search quality benchmarkSlots into the search stack you have
Add Rerank as a final ranking pass — no migration, no re-indexing, just sharper results.
Works with any retriever
Sit Rerank after keyword search, a vector database, or a hybrid pipeline — it takes a candidate list and returns it reordered, no rebuild required.
Simple API & SDKs
Send a query and your documents to one endpoint, get back scored, ranked results. Client libraries cover the languages your stack already speaks.
Tunable top-k & scoring
Control how many candidates to re-score and how results are returned — balance precision, latency, and cost for each search surface.
“We added Rerank as one extra call after our vector search, and the answer our users wanted started showing up first. It was the smallest change with the biggest jump in search quality we have shipped.”
MERIDIAN
Renata Voss
VP, Search Engineering
Cross-encoder relevance
Rerank scores the query and each document jointly, capturing meaning a bi-encoder retriever alone can miss.
Long-context passages
Score long documents in a single pass and surface the exact section that answers the question.
Multilingual ranking
Reorder results across many languages, including queries and documents that do not share one.
Tunable top-k
Choose how many candidates to re-score per request to balance precision against latency and cost.
Calibrated scores
Every result comes back with a relevance score you can threshold, filter, or audit downstream.
Low-latency inference
Re-rank a full candidate set in tens of milliseconds — fast enough to sit in a live search path.
BREVA
How a help center cut support tickets with sharper search
A global appliance maker added Rerank to its documentation search. With the right article surfacing first, customers solved more problems on their own.
Resources to get started
Everything you need to evaluate and integrate Rerank.