US stock futures dip as Trump’s firing of Cook sparks Fed independence fears
Investing.com -- Alibaba (NYSE:BABA) has launched a new artificial intelligence model, Qwen2.5-VL-32B-Instruct, an optimized version of the Qwen2.5-VL series that was rolled out at the end of January this year. The new model, which is released under the Apache 2.0 license, has been designed to provide more detailed, well-structured answers that align more closely with human preferences.
The Qwen2.5-VL-32B-Instruct model boasts significant improvements in mathematical reasoning accuracy. It also exhibits enhanced capabilities in fine-grained image understanding and reasoning, including tasks such as image parsing, content recognition, and visual logic deduction.
The new model has demonstrated superior performance over comparable state-of-the-art models, including Mistral-Small-3.1-24B and Gemma-3-27B-IT. It also outperforms the larger Qwen2-VL-72B-Instruct model, particularly in multimodal tasks such as MMMU, MMMU-Pro, and MathVista, which require complex, multi-step reasoning. On the MM-MT-Bench, a user experience evaluation benchmark, Qwen2.5-VL-32B-Instruct significantly outperformed its predecessor.
In addition to its visual capabilities, Qwen2.5-VL-32B-Instruct has achieved top-level performance in pure text capabilities at the same scale.
Alibaba’s future research direction will focus on improving long and effective reasoning processes. This strategic move aims to enhance the performance of visual models in handling highly complex, multi-step visual reasoning tasks.
This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.