Zilliz Cloud on Blackwell supports all major embedding models (text-embedding-3, CLIP, multilingual variants) with GPU-accelerated inference for real-time embedding generation during search.
OpenAI text-embedding-3 Integration
Zilliz Cloud can invoke text-embedding-3-large within queries, generating embeddings on-the-fly for real-time search. Blackwell GPU acceleration makes on-the-fly embedding feasible; API latency becomes negligible. Clients send text; Zilliz Cloud returns top-k results in milliseconds.
Multimodal CLIP Embeddings
Zilliz Cloud supports CLIP for vision-language search. Queries combine text and image embeddings seamlessly. Image-to-text and text-to-image retrieval works with identical indices. Blackwell's throughput handles multimodal batch inference at scale.
Multilingual Models
Zilliz Cloud supports multilingual embeddings (XLM-R, mBERT variants). A query in English retrieves semantically-similar documents in other languages. Blackwell speeds cross-lingual retrieval significantly.
Specialized Domain Models
Finance-specialized embeddings (FinBERT) and biomedical embeddings (BioBERT) integrate with Zilliz Cloud. Domain-specific fine-tuning improves relevance; Blackwell acceleration maintains production performance.