Alibaba shares rose about 2% on Monday morning after the company launched a new artificial intelligence model designed to run on consumer hardware, intensifying competition with Meta’s recent open-source push.

Alibaba Group Holding Limited, BABA

The new model, Qwen3.8-27B, is built to run on a single consumer graphics card or a high-end laptop in its 4-bit quantized form, using just 17 GB of memory. Alibaba said the model outperforms systems many times its size and rivals OpenAI’s Claude Opus 4.6 on agentic tasks.

The release came after Meta announced last week that it would open-source its own AI model, Muse Glimmer, which is also designed to run on laptops. Meta described that move as a U.S. alternative to Chinese AI technology.

Alibaba also released the weights for Qwen3.8 Max, its most powerful model. Releasing weights allows developers to freely download, run, and fine-tune the model, forming the basis of the open-weight AI ecosystem, where developer adoption and model accessibility can matter as much as headline performance.

Qwen3.8-27B replaces the earlier Qwen3.6-27B. The new model is licensed under Apache 2.0, which permits commercial use and modification.

Downloads Tell the Story

Within three days of its August 14 launch, Qwen3.8-27B surpassed 3 million downloads on Hugging Face. That pace reflects broader momentum around Alibaba’s model family: Qwen-based models now account for 151,448 derivative models on the platform, 2.6 times Meta’s total footprint.

“The company that can offer the most capable open weights models will move ahead in this race,” said Neil Shah, co-founder at Counterpoint Research.

Alibaba has been a leader in open-weight AI for some time, while other Chinese companies such as DeepSeek and Moonshot have also been active in the field. Meta was an early entrant in open-source AI with its Llama family, but Chinese developers quickly overtook it.

“Meta’s own re-embrace of open weights was itself a response to two years of Chinese labs taking a large share of the open-weight market,” said Nick Patience, AI lead at Futurum Group.

On-Device AI Becomes the Next Battleground

The launch of Qwen3.8-27B points to a broader shift in the industry. Running AI models locally on devices rather than in data centers can improve speed and privacy, and analysts say on-device AI is becoming a major competitive arena as companies look to make models usable outside cloud infrastructure.

Shah described on-device AI as the “next battleground” for AI developers. Patience said Alibaba has built an advantage across both open-weight and on-device AI categories.

“Alibaba has made Qwen the most credible non-U.S. model family to build hardware relationships around, in China and in the open-weight developer community globally,” Patience said.

The model has not been without criticism. Some observers say its dense architecture makes it slower than systems that use a mixture-of-experts approach. Even so, its ability to run on a single GPU has attracted attention from developers experimenting with local deployment.

“The fact that a 17 GB file can do all of this stuff on my home machines is a miracle,” UK software engineer Simon Willison posted on Reddit.

BABA stock was up about 0.38% as of Monday’s session, after earlier gains of roughly 2% at the open.