Zhipu Embedding 3
Awaiting activationFind Chinese and English business knowledge with vectors sized for your index.
What it does
Embedding 3 is Zhipu AI’s third-generation text vector model. Its official release describes improved Chinese and English semantic representation and adaptation to terminology in technology, finance and medical research. Those are vendor claims to evaluate against your actual collection, rather than a guarantee for every domain.
The default output has 2,048 dimensions and can be reduced. The documentation’s overview says 8K context, while its request specification limits each array entry to 3,072 tokens; this integration follows the stricter request limit and a 3,000-byte service cap.
Language coverage
Zhipu explicitly names Chinese and English semantic retrieval. Its broader multilingual claim does not establish a tested list of other languages.
Official language coverage ↗Where it helps
Chinese knowledge bases
Find passages answering customer or employee questions.
Domain terminology
Evaluate retrieval on finance, technical and research vocabulary.
Content organization
Group related Chinese and English feedback or articles by meaning.
When to choose it
Consider it when Chinese content and specialist vocabulary are central to your market. Evaluate the retrieval result on real documents before treating it as a domain specialist.
Awaiting activation. This is a general text retrieval model with provider-reported domain adaptation, rather than a clinical decision model.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
Zhipu AI · Embedding 3 model release and API examples ↗Reviewed October 4, 2026