Text and Multimodal Embeddings

Protocol selection for text-embedding-v4, Gemini Embedding and Qwen3 VL Embedding.

Text embeddings use POST /v1/embeddings:

{"model":"text-embedding-v4","input":["A quiet garden","A busy street"]}

Read data[].embedding. Gemini models gemini-embedding-001 and gemini-embedding-2 use POST /v1beta/models/{model}:embedContent:

{"content":{"parts":[{"text":"A quiet garden"}]}}

Use your ModelSell Bearer key and read embedding.values. Text compatibility depends on the channel; media support requires the native content format.

qwen3-vl-embedding uses native input.contents for multimodal input. Its Alibaba native route is not registered in the current gateway. Catalog availability alone does not establish native media support; confirm the channel before sending media.

Embedding models do not accept ordinary chat messages or :generateContent.

Sources checked 2026-10-04: Google, Alibaba embeddings, multimodal API.