VoyageAI Launches Advanced Embedding Models with Superior Performance and Cost-Effectiveness
TapTechNews September 28th news, the tech media marktechpost released a blog post yesterday (September 27th), reporting that VoyageAI company has launched two Embedding models, Voyage-3 and Voyage-3-Lite.
Both of these models show advantages over similar models in aspects such as technology, law, finance, multilingual applications, and long-text understanding. While maintaining a relatively small model size and low resource consumption, they provide developers with more efficient and easier-to-integrate solutions.
TapTechNews cited official data from VoyageAI. In terms of technical documents, code, law, finance, web content, multilingual data sets, long documents, and conversation data, Voyage-3 significantly reduces the cost of the vector DB with a 1/2.2 lower cost and a 1/3 smaller embedding dimension, and its comprehensive跑分 exceeds OpenAI's v3Large model by 7.55%.
Voyage-3-Lite has a 3.82% higher retrieval accuracy than OpenAI's v3Large model, while the cost is reduced to 1/6 and the embedding dimension is also reduced to 1/6.
Embedded AI models
An Embedding model is a machine learning model widely used in fields such as natural language processing (NLP) and computer vision (CV). It can convert high-dimensional data into a low-dimensional embedding space and retain the features and semantic information of the original data, thereby improving the efficiency and accuracy of the model.
Voyage-3 doesn't compromise on quality and improves cost-effectiveness
Cost-effectiveness is the core of the new Voyage-3 series models.
Voyage-3 has a context length of 32,000 tokens, which is 4 times that of OpenAI products. It is a cost-effective solution for enterprises that require high-quality retrieval, and the price is affordable.
Voyage-3 costs $0.06 per million tokens, which is 1.6 times cheaper than CohereEnglishV3 and much more affordable than OpenAI's v3Large model.
Voyage-3-Lite, the lightweight version of this model, is optimized for low-latency operations. The cost per million tokens is $0.02, which is 1/6.5 of OpenAI's v3Large model, and the embedding dimension is 1/6 to 1/8 (512 vs. 3072 of OpenAI).
VoyageAI
VoyageAI is an AI company founded by assistant professor Ma Tengyu from Stanford University in November 2023, focusing on building Embedding/vectorized models to improve the efficiency of data processing.
The company consists of multiple AI researchers, including Stanford University professor Ma Tengyu and MIT doctor, and has received academic advisory support from Christopher Manning, the director of the Stanford Artificial Intelligence Laboratory, and famous Chinese scholar in the AI field Li Feifei, etc.
The API endpoints provided by VoyageAI can receive user data and return embedding s or correlation scores, and these models can be seamlessly integrated with other parts of the RAG stack, including vector storage and generative large language models (LLMs).