Alibaba Cloud has introduced the latest variant of its flagship model, Qwen 3.8-27B, in another sign that open model competition continues to intensify between Chinese and U.S. model makers.

Released over the weekend under the Apache 2.0 license, the open-weight model is optimized for software engineering, reasoning and long-horizon tasks. Alibaba said the model can provide stronger autonomous planning and flexible control over reasoning, allowing users to determine how deeply the model should reason and whether the context used to support that reasoning is retained or deleted.

The release comes a few days after U.S. vendors Meta and Nvidia introduced their own new open models. It also highlights the continuing push to bring AI to edge devices, where local execution can matter for organizations balancing performance, privacy and cost. Like Meta Muse Glimmer and Nvidia’s Nemotron 3.5 Lightning, the smaller 27B variant of Qwen is designed to run locally on a laptop. Alibaba’s most powerful model, Qwen 3.8-Max, has 2.4 trillion parameters and is 88 times larger than Qwen 3.8.

The move toward edge deployments reflects the growing attention vendors are paying to enterprise AI costs and efforts to reduce them.

“Many companies are looking to gain a better understanding of their token spend, what they’re actually getting out of the model, the value proposition,” said Bradley Shimmin, an analyst at Futurum Group.

Edge is About Cost

Shimmin said many enterprises find it difficult to understand the cost of accessing frontier models through APIs. Running models locally, he said, gives organizations greater control over the cost of model operations.

“It’s really a way of applying FinOps, the idea of operationalizing the finances of AI,” Shimmin said. “Because of this renewed interest in observability and FinOps, we’re seeing these companies finetuning their spend.”

Alibaba is also targeting the edge AI market because of its partnership with Apple. Alibaba will bring AI features to Apple devices sold in China, and a model that can run on edge devices helps Alibaba reach Apple users in the country.

“All the Apple users in China will probably be using Qwen instead of Gemini, so there is a benefit in Qwen targeting this market because it has a large installed base,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.

The AI Open Source War

Alibaba’s competition with Meta in open models is not surprising, Su said, given the success of Meta’s earlier open models, including the Llama family, which became one of the most popular open model lines after its release in 2023.

“There is an argument to be made that Meta can really do good open weight models,” Su said. “For Alibaba, Qwen is a very important way for them to market themselves. They want to continue to carry the open source mantle.”

Shimmin said Alibaba and other Chinese vendors have become major drivers of open source over the past year.

He added that the release of multiple Chinese open models helped prompt Meta to reverse course on its policy of releasing only open models, restoring a balance in the AI market between open models and closed, proprietary models, a segment long dominated by U.S. AI labs Anthropic and OpenAI.

“The AI industry was built on the back of open source technologies,” Shimmin said. “The ideal of open source has an ecosystem-wide collaborative effort, not something that was locked down with IP containers opaque to users trying to understand, even the basics of why a model says what it does or how their spending is being utilized.”