Jina EmbeddingsUnclaimed

Jina AI's Embeddings API converts text, images, audio and video into vectors, numerical representations that can be compared for meaning. The page describes its models as built for search, RAG and agent applications. The current v5-omni models place all four types of input in one shared embedding space, while v5-text is a small multilingual text model with a 32K context window on the small size and 8K on the nano size.

Developers send a POST request to api.jina.ai/v1/embeddings with an API key as a bearer token, passing text or images in a JSON body. Output can be returned as float, binary or base64, and vectors can be normalised for cosine similarity. The page lists integrations with vector stores such as MongoDB, Qdrant, Pinecone and Weaviate, and with frameworks such as LlamaIndex, Haystack and LangChain.

The API is aimed at teams building search and retrieval-augmented generation systems. Businesses can also deploy the models on AWS SageMaker or Microsoft Azure, or ask Jina AI for a Kubernetes deployment in a virtual private cloud or on their own servers.

Access starts with a free trial and an API key. Usage is paid for by topping up tokens, with an option for automatic top-up when the balance runs low. Rate limits are set per key and per tier, counted in requests and tokens per minute. The page does not give a fixed monthly price for the embeddings endpoint, and says its cost depends on the size of the input.