voyage-large-2 can support summarization workflows well, but it does not generate summaries by itself because it is an embedding model, not a text generator. What it gives you is a strong way to represent long text as vectors, which helps you select the right content to summarize (for example, the most relevant passages, the most representative chunks, or the most diverse set of topics). The Zilliz model guide even calls out voyage-large-2 as “ideal for tasks like summarization,” which is best understood as “useful for building summarization pipelines,” not “a summarizer endpoint.” In short: voyage-large-2 improves the retrieval and organization steps that make summaries better and more consistent.
A concrete pattern is “retrieve-then-summarize.” Suppose you want a weekly summary of 10,000 support tickets. You chunk each ticket (or treat each ticket as a unit), embed the text with voyage-large-2, and then do one of two things: (1) cluster embeddings to group similar issues and summarize each cluster; or (2) retrieve the most representative tickets for a theme and summarize those. For clustering, you might run k-means on vectors and then pick the nearest tickets to each centroid as exemplars. For retrieval, you can embed a theme query like “billing failures after upgrade” and fetch top-k nearest neighbors, then summarize only those results. This keeps summarization focused and avoids the “I tried to summarize everything and got mush” problem that happens when you feed too much mixed content into a generator.
This is where a vector database such as Milvus or Zilliz Cloud (managed Milvus) becomes the practical companion to voyage-large-2. You store each chunk embedding plus metadata (ticket_id, timestamp, product, language, severity). Then you can do fast similarity search (“give me the top 200 most central tickets this week”) or filtered search (“only severity=SEV1, only product=X”), and your summarization step runs on a curated subset. In other words, voyage-large-2 helps you decide what to summarize, and Milvus/Zilliz Cloud helps you do that selection quickly and repeatably at scale.
For more information, click here: https://zilliz.com/ai-models/voyage-large-2