Alibaba Cloud Bailian Platform has launched the Qwen2.5-Turbo million long text model. This model was developed by the Tongyi Qianwen team and has the ability to process ultra-long text of up to 1 million tokens, which is equivalent to 1 million English words or 150 Ten thousand Chinese characters. This model has achieved excellent results in long text retrieval, long text evaluation set RULER, and long text tasks close to real scenes, surpassing GPT-4 in multiple dimensions. In addition, it also performed well in the short text benchmark test, significantly surpassing previous similar models.
Alibaba Cloud Bailian Platform recently announced the launch of the Qwen2.5-Turbo million long text model. The Qwen2.5-Turbo model was developed by the Tongyi Qianwen team and supports processing of ultra-long contexts of up to 1 million tokens, which is equivalent to 1 million. English words or 1.5 million Chinese characters.
This new version of the model achieved 100% accuracy in long text retrieval tasks, and scored 93.1 on the long text evaluation set RULER, surpassing GPT-4. In long text tasks close to real scenes such as LV-Eval and LongBench-Chat, Qwen2.5-Turbo surpasses GPT-4o-mini in most dimensions. In the short text benchmark test, Qwen2.5-Turbo also performed very well, significantly surpassing the previous open source model with a context length of 1M tokens.
The Qwen2.5-Turbo model has a wide range of application scenarios, including in-depth understanding of novels, large-scale code assistants, reading of multiple papers, etc. It can process 10 novels, 150 hours of speeches, or 30,000 lines of code at one time. In terms of reasoning speed, the Tongyi Qianwen team compressed the calculation amount by about 12.5 times through the sparse attention mechanism, and reduced the first word return time of processing 1M tokens context from 4.9 minutes to 68 seconds, achieving a 4.3 times speed increase.
Alibaba Cloud Bailian platform provides all users with the ability to directly call Qwen2.5-Turbo API, and provides a limited-time gift of 10 million tokens. The cost of subsequent use of one million tokens is only 0.3 yuan.
At present, Alibaba Cloud Bailian platform has launched more than 200 domestic and foreign mainstream open source and closed source large models, including Qwen, Llama, and ChatGLM, supporting users to directly call, train and fine-tune or create RAG applications.
The emergence of the Qwen2.5-Turbo model marks significant progress in long text processing technology, providing more powerful AI tools for all walks of life. The open strategy of Alibaba Cloud Bailian platform also allows more developers to easily apply this advanced technology and jointly promote the development of the field of artificial intelligence. Its low cost also further reduces the threshold for use.