Google Releases Gemini Embedding Model

Gemini Embedding utilizes Matryoshka representation learning, which allows developers to adjust the output dimension based on their specific needs (default is 3072, but can be reduced to 1536 or 768), offering a balance between performance and storage cost.

Additionally, the model is available in both free and paid tiers, with pricing set at $0.15 per 1 million input tokens.