This page lists common embedding models and their maximum input length (in tokens), for filling in the knowledge base settings. Providers update their models often — check the linked official pages for the latest list.
To stay on the safe side, some max input values here are slightly below the official maximum (for example 8191 instead of 8192). If you're unsure, use the values in this table.
Want to work offline? BLACK HOLE also has a built-in local embedding model (Qwen3 Embedding 0.6B) — see Local Models .
OpenAI
Official model information
Name max input text-embedding-3-small 8191 text-embedding-3-large 8191
Google Gemini
Official model information
Name max input gemini-embedding-001 2048
Jina
Official model information
Name max input jina-embeddings-v4 32768 jina-embeddings-v3 8191 jina-clip-v2 8191
SiliconFlow
Official model information
Name max input Qwen/Qwen3-Embedding-8B 32768 Qwen/Qwen3-Embedding-4B 32768 Qwen/Qwen3-Embedding-0.6B 32768 BAAI/bge-m3 8191 netease-youdao/bce-embedding-base_v1 512
Cohere
Official model information
Name max input embed-v4.0 128000 embed-english-v3.0 512 embed-multilingual-v3.0 512
Together AI
Official model information
Name max input M2-BERT-80M-2K-Retrieval 2048 M2-BERT-80M-8K-Retrieval 8192 M2-BERT-80M-32K-Retrieval 32768 UAE-Large-v1 512 BGE-Large-EN-v1.5 512 BGE-Base-EN-v1.5 512
nomic
Official model information
Name max input nomic-embed-text-v1 8192 nomic-embed-text-v1.5 8192 gte-multilingual-base 8192
Upstage
Official model information
Name max input embedding-query 4000 embedding-passage 4000
