We use cookies. Cookies and browser storage keep you signed in and remember your preferences. With your permission we also measure how the site is used. You can any time. See our cookie policy.
mongodb/voyage-4-liteRuns on your balance at the prices above. New accounts start with free usage.
Settings
VoyageAI by MongoDB: voyage-4-lite is a lightweight, general-purpose embedding model from VoyageAI by MongoDB. It sits in the embeddings category, so its job is to turn text into vector representations rather than to generate chat responses. The model is built for low latency and cost, which makes it a reasonable default when you need embeddings at volume and do not want a heavier model in the request path. It was released by the provider on 2026-07-27.
The model accepts text as input and produces embeddings as output. Its context length is 32000 tokens, and the maximum output is 28800. The provider notes that voyage-4-lite is enabled by Matryoshka learning and quantization-aware training. Matryoshka learning is relevant to developers who want to truncate embeddings to shorter dimensions while keeping useful structure, and quantization-aware training is relevant when embeddings need to stay accurate under compressed representations. Both matter when you are choosing an embedding model for a retrieval system rather than for a single one-off vector.
Practical use cases follow directly from those properties. Semantic search and retrieval over document collections are the obvious fit, including vector search inside MongoDB Atlas and similar stores. Because the model is general purpose, it also works for clustering and topic grouping, deduplication and near-duplicate detection, classification features built on top of embedding vectors, and recommendation or similarity matching where items are compared by meaning rather than by keyword overlap. The 32000 token context means longer passages, such as full pages or multi-section documents, can be embedded in a single request instead of being split into many small chunks, which simplifies indexing pipelines.
You call the model through api14's OpenAI-compatible API using the model id "mongodb/voyage-4-lite". Because the endpoint follows the OpenAI embeddings shape, you send your text in the input field and read the vectors back from the response. The model supports no additional request parameters beyond the standard embeddings call, so there is no temperature, top_p or similar sampling configuration to set. Keep requests within the 32000 token context and expect outputs up to 28800.
If you need a model that generates text, handles images or audio, or supports tunable generation parameters, an embeddings model is the wrong tool and a chat or multimodal model will fit better. Within embeddings work, voyage-4-lite is aimed at low latency and cost, so if your task depends on a larger or more specialized embedding model, consider what else the provider offers.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.embeddings.create({
"model": "mongodb/voyage-4-lite",
"input": [
"The quick brown fox",
"Lorem ipsum"
]
});
console.log(result.data.map((d) => d.embedding.length));
This community has no topics yet. Open the first one about VoyageAI by MongoDB: voyage-4-lite.