Embedding Models

BGE-EN-ICL is BAAI’s English embedding model, which can take a few examples in the query to adapt to a task, an approach BAAI calls in-context learning.

BGE-M3 is BAAI’s open-weight multilingual embedding model, covering more than 100 languages. It turns inputs of up to 8,192 tokens into 1,024-dimensional vectors for semantic search, retrieval and clustering across languages.

Gemini Embedding 2 Preview$0.25/ 1M tok

Gemini Embedding 2 Preview is a preview of Google’s Gemini-based embedding model, returning 3,072-dimensional vectors for inputs of up to 2,048 tokens. It’s built for semantic search, retrieval and classification.

Multilingual E5 Large Instruct$0.0125/ 1M tok

Multilingual E5 Large Instruct is an open-weight embedding model from Microsoft’s E5 family, covering about 100 languages.

Nemotron Embed VL 1B v2$0.0125/ 1M tok

Nemotron Embed VL 1B v2 is NVIDIA’s open-weight embedding model from its Llama Nemotron family, trained for retrieval over visually rich documents such as reports, slides and scanned pages.

Qwen3 Embedding 0.6B$0.0125/ 1M tok

Qwen3 Embedding 0.6B is the smallest model in Alibaba’s open-weight Qwen3 Embedding series, a fast, low-cost choice for multilingual and code retrieval. It returns 1,024-dimensional vectors.

Qwen3 Embedding 8B$0.0125/ 1M tok

Qwen3 Embedding 8B is the largest model in Alibaba’s open-weight Qwen3 Embedding series, built on Qwen3 and trained for more than 100 languages, including programming languages.

Text Embedding 3 Large is OpenAI’s larger embedding model, returning 3,072-dimensional vectors for inputs of up to 8,191 tokens.

Text Embedding 3 Small is OpenAI’s efficient embedding model, returning 1,536-dimensional vectors for inputs of up to 8,191 tokens.