Voyage Rerank 3 and Rerank 3 Lite are now available on AI Gateway
By Steven Van ·
Both models support 32K tokens per query-document pair, with Rerank 3 focused on accuracy and Lite optimized for latency and cost.
Search results can now be ranked for relevance before documents reach a language model using Voyage AI by MongoDB’s latest rerankers on Vercel AI Gateway. The release note reports retrieval improvements across domains, with the largest gains on long documents and code.
Accuracy or lower latency
Rerank 3 focuses on accuracy, while Rerank 3 Lite is optimized for latency and cost. Both support up to 32K tokens per query-document pair and follow instructions in the query to guide ranking.
Ranking documents
Call the AI SDK’s rerank function with a query, a list of documents and the model identifier voyage/rerank-3. Switch to voyage/rerank-3-lite to use Lite. The release note’s example sets topN to 1 to return the highest-ranked document.
Pricing and access
The Rerank 3 model page lists input pricing at $0.05 per million tokens. Developers can use an AI Gateway API key or their own provider key, and try Rerank 3 in the AI Gateway model playground.