适用于 AI Agent 的 Jina AI API

Glasser lists document reranking and two text embedding operations: jina-embeddings-v3 and jina-embeddings-v5-text-small. Both embedding operations use native late chunking.

Embeddings convert supplied text into vectors your application can index for similarity search. Reranking takes a query and supplied documents, then scores their relevance. Your application can use reranking after retrieving a candidate set.

The two listed embedding operations have late_chunking enabled. All texts supplied in a request share the model’s context window, and the provider rejects input that exceeds it. Inspect the selected endpoint’s current input requirements before preparing a batch.

These three operations process text or documents you supply. For source discovery or webpage extraction, choose a search or extraction provider first, then pass the text to the relevant Jina operation in your own workflow.

The listed embedding operations generate vectors. Your application handles vector storage, indexing and retrieval. The reranker separately scores supplied documents against a query.

Calls through Glasser use your Glasser key and workspace balance. Direct Jina access uses a Jina API key and token balance, with separate free and paid rate limits. Jina offers token top-ups for its hosted APIs.

Use the live Glasser endpoint directory for the current quote, billing unit and cap, then inspect the selected operation for its full price. Direct Jina embedding and reranking usage is counted from input tokens; the direct dashboard manages its token balance and top-ups.