VoyageAI by MongoDB: voyage-multimodal-3.5

VoyageAI by MongoDB: voyage-multimodal-3.5 is a multimodal embedding model from Voyage AI, a company now part of MongoDB. It sits in the embeddings category rather than the chat category, so its job is to turn input content into vectors rather than to generate text responses. The model was released by the provider on 2026-07-27.

The model accepts text and image inputs and produces embeddings as output. Its context window is 32,000 tokens, and the maximum output is 28,800. According to the provider description, voyage-multimodal-3.5 is capable of vectorizing text, images and video individually, and also content that interleaves all three modalities, which makes it a general purpose embedding model rather than a text only one.

That combination of capabilities points to a few practical uses. Because it handles both text and images in one embedding space, it fits search and retrieval systems where queries and documents may not be the same shape, for example a text query matched against an image catalogue, or mixed media documents embedded as a single unit. Interleaved content support means pages that combine prose, pictures and video in sequence can be vectorized without splitting them into separate pipelines. Clustering, deduplication, recommendation and classification over mixed media collections are natural fits as well.

You call it through api14's OpenAI-compatible API using the model id voyageai/voyage-multimodal-3.5, which means existing OpenAI client libraries and request patterns carry over with the model name swapped. In terms of request parameters, the model lists no supported parameters, so there is no tuning surface to configure in the request body beyond the model id and the input content itself. You send the text and image input, and you get embeddings back.

Because the supported parameter list is empty, if you need control over dimensions, truncation behaviour or other embedding options, check whether another embedding model in api14's catalogue exposes those knobs before committing. Similarly, this model does not generate text, so tasks that need summarisation, question answering or chat should go to a generative model instead. Choose voyage-multimodal-3.5 when the task is representation rather than generation, and when your content mixes text and images.