VoyageAI by MongoDB: voyage-4

voyage-4 is a general-purpose embedding model from Voyage AI, the company behind the Voyage embedding family that is now part of MongoDB. It is listed on api14 under the model id voyageai/voyage-4 and sits in the embeddings category, so it is built to turn text into vectors rather than to generate chat responses. If you are choosing an embedding model for a retrieval or search system, this is the kind of model you would evaluate alongside other general-purpose embedding options.

The model accepts text as input and returns embeddings as output. It is described as general-purpose and multilingual, and it is optimized for retrieval and search as well as other AI applications. It supports embeddings at 2048, 1024, 512, and 256 dimensions, along with multiple quantization options, which lets you trade vector size against storage and search characteristics depending on your index. The context length is 32000 tokens, and the maximum output is 28800, so long documents can be embedded without splitting them into many small pieces.

Those capabilities point to a fairly specific set of uses. Retrieval and search are the primary ones: indexing a document collection, embedding queries, and comparing them by vector similarity. Because the model is multilingual, it also fits collections and queries that mix languages, and because it supports several output dimensions and quantization options, it can serve both a high-fidelity index and a smaller, more compact one. Common applications include retrieval augmented generation pipelines, semantic search over internal documentation, deduplication or clustering of text, and recommendation or classification systems that operate on embeddings rather than generated text.

You call it through api14's OpenAI-compatible API using the model id voyageai/voyage-4, sending text and receiving embeddings in the response. The model's supported parameters list is empty, so you should not expect to pass sampling controls such as temperature or top_p; the request is essentially the input text plus the model id. Dimension and quantization choices are part of the model's embedding behavior rather than request parameters exposed here, so check the current API reference for how those are selected before you build around a specific vector size.

One thing to keep in mind is that this is an embedding model, not a chat or completion model. If your task is generating text, summarizing, or holding a conversation, a text generation model is the right fit, and voyage-4 would only be useful as a supporting component in a retrieval pipeline. It also takes text only, so inputs in other modalities are out of scope. Within those bounds, it is a reasonable candidate for search and retrieval work where you want control over embedding dimensions.