<div id="readability-page-1"><div><div><p><strong>Alibaba Group</strong> <span>BABA</span> launched an upgraded version of its third-generation Qwen3 family of large language models (LLMs), enhancing its AI capabilities in math and coding to outperform top models from OpenAI and DeepSeek in key benchmarks.</p><p>The new open-source model, Qwen3-235B-A22B-Instruct-2507-FP8, delivered marked improvements in instruction following, logical reasoning, coding, science, and tool usage, as noted in updates on HuggingFace and Alibaba’s ModelScope platform, SCMP reported on Tuesday. </p><p>While the upgraded Qwen model currently operates in a non-thinking mode, delivering direct outputs without displaying reasoning steps, it now supports inputs of up to 256,000 tokens. This represents an eightfold increase, enabling the model to process much longer text sequences within a single conversation.</p><p><strong><em>Also Read: Alibaba, Huawei Dominate China’s Growing Cloud Market</em></strong></p><p>Beyond its core LLM advancements, Alibaba also announced that its 3-billion-parameter Qwen model will power HP’s “Xiaowei Hui” smart assistant in China. This integration aims to help users more efficiently draft documents and summarize meetings.</p><p>Performance data released by the company shows Qwen3-Coder surpassed domestic rivals such as DeepSeek and Moonshot AI’s K2 in key coding benchmarks. Alibaba also said the model rivals top U.S. offerings, including OpenAI’s GPT-4 and Anthropic’s Claude, in select capabilities.</p><div><div><p>Trending Investment Opportunities</p></div></div><p>During a recent visit to China, <strong>Nvidia </strong><span>NVDA</span> CEO <strong>Jensen Huang</strong> praised the Qwen series, alongside DeepSeek and Moonshot’s Kimi models, as among the world’s most advanced open reasoning AI systems, underscoring China’s rapid progress in the global AI race.</p><p>Huang visited China shortly after the U.S. approved export license filings for the company’s H20 GPUs, which were designed to meet American export regulations. Although U.S. agencies have not finalized the paperwork, Huang said the government assured Nvidia that licenses would be granted.</p><p>Alibaba stock has seen substantial gains this year, increasing over 42% year-to-date and outperforming the NYSE Composite Index’s rise of over 8%. </p><p><strong>Price Action: </strong>BABA shares were trading higher by 2.15% to $123.31 premarket at last check Wednesday.</p><ul> <li><strong><em>Alibaba Bleeds Billions As Delivery Price War Erases $100 Billion In Market Value</em></strong></li> </ul><p><em>Photo by Mamun_Sheikh via Shutterstock</em></p></div></div></div>

BABA

09988

KBAB

BABO

BABX

<p>Alibaba Group has launched an upgraded version of its Qwen3 large language models, outperforming OpenAI and DeepSeek in key benchmarks. The new model, Qwen3-235B-A22B-Instruct-2507-FP8, shows significant improvements in instruction following and coding, with an increased input capacity of 256,000 tokens. Additionally, Alibaba's 3-billion-parameter Qwen model will power HP’s smart assistant in China. Nvidia's CEO praised the Qwen series during a recent visit to China, highlighting its advanced capabilities. Alibaba's stock has risen over 42% this year, outperforming the NYSE Composite Index.</p>

- Alibaba Group launched an upgraded version of its Qwen3 family of large language models, enhancing AI capabilities in math and coding to outperform competitors like OpenAI and DeepSeek.  
- The new model, Qwen3-235B-A22B-Instruct-2507-FP8, supports inputs of up to 256,000 tokens and integrates with HP’s smart assistant in China.  
- Performance data indicates that Qwen3-Coder surpasses domestic rivals and rivals U.S. models, contributing to Alibaba's stock increase of over 42% year-to-date.

Alibaba's Upgraded Qwen3 Outperforms OpenAI, DeepSeek