voyage-3-large: the new state-of-the-art general-purpose embedding model

Voyage-3-large is a cutting-edge, general-purpose, multilingual embedding model that leads in performance across eight domains and 100 datasets, including law, finance, and code. It surpasses OpenAI-v3-large and Cohere-v3-English by 9.74% and 20.71% on average, respectively, and is enabled by Matryoshka learning and quantization-aware training, which support smaller dimensions and quantization options to significantly reduce vectorDB costs while maintaining retrieval quality. The model offers various embedding precisions, including 32-bit, int8, and binary, with a 32K-token context length, enabling it to balance retrieval quality with storage efficiency. This model establishes a new accuracy-cost frontier, outperforming previous Voyage models and OpenAI-v3-large with reduced storage needs, and offers improvements in retrieval quality when used with binary rescoring. Voyage-3-large is now available, with the first 200 million tokens offered for free, and further information can be accessed via their documentation or through their social media and contact platforms.