Alibaba unveils new AI embedding models a field it leads globally
4 days ago
Alibaba Group Holding has made its Qwen3 Embedding series available for developers in the Chinese tech giants latest bid to solidify its global leadership in opensource artificial intelligence AI models
Released late on Thursday the series marks another addition to the companys lineup of large language models LLMs which are among the worlds most popular opensource AI systems according to New Yorkbased computer app company Hugging Face
Alibaba owner of the South China Morning Post ranks third globally in the field of LLMs according to the 2025 AI Index Report from Stanford University
The new models which come in various parameters support over 100 languages including multiple programming languages and provide robust multilingual crosslingual and code retrieval capabilities according to Alibaba
In AI an embedding model helps computers understand and process text by turning it into numerical representations Since computers process data solely in numerical form the embedding process enables them to grasp semantic data and questions more effectively delivering more tailored results that do not rely solely on keywords
Alibaba holds the top position on the Massive Text Embedding Benchmark a ranking published by New Yorkbased computer app company Hugging Face that measures the performance of textembedding models
Hangzhoubased Alibaba said the new series would enable ongoing optimisation of the Qwen foundation model resulting in enhanced training and improved efficiency of its embedding and reranking systems The reranking process refines the order of search results to better match a users query
The new model follows the same multistage training paradigm used in previous models from the companys general textembedding series according to the announcement
This threestage training process involves an initial contrastive examination of large quantities of raw data to assess the systems capacity to separate data based on relevance The second stage tests this process with higherquality curated data while the third stage combines these findings to enhance overall performance
Alibaba described the new Qwen3 series as a new starting point and said it was excited to see more developers implement its product in diverse scenarios