Alibaba has introduced the Qwen2.5-Turbo model, an upgraded artificial intelligence model based on the Open Source version. This model focuses on data capacity, increasing from 128,000 tokens to 1 million tokens, enabling it to support book-level data, equivalent to approximately 1 million Chinese characters or 1.5 million letters.
The testing results of Qwen2.5-Turbo are quite impressive, scoring a perfect 100 on Passkey Retrieval, a significant improvement over GPT-4. When inputting a million tokens, it responds in just 68 seconds at a rate of 0.3 yuan per million tokens, significantly cheaper than GPT-4o-mini at 4.6 times the cost.
Currently, this model is available on Alibaba Cloud Model Studio and can be tested for free on HuggingFace.
TLDR: Alibaba unveils Qwen2.5-Turbo model with enhanced data capacity and improved performance, available for use on Alibaba Cloud Model Studio or for free testing on HuggingFace.
Leave a Comment