Alibaba Launches Qwen4-Max AI Model

The closed model, with 2.4 trillion parameters, arrives as Alibaba shares slipped more than 1% in overnight trading.

Updated on Nov. 16, 2026 in Artificial Intelligence

Vector editorial illustration of a row of plain server cabinets with a single coral cooling duct, rendered in flat teal, cream, and slate.
Alibaba released Qwen4-Max, a closed AI model with 2.4 trillion parameters, as its shares slipped more than 1% in overnight trading. AI Illustration. Upload story photo >

Alibaba has launched Qwen4-Max, its most advanced AI model, which is a closed model comprising 2.4 trillion parameters. Alibaba shares fell more than 1% in the overnight session late Sunday.

Why it matters

Qwen4-Max delivers stronger reasoning while activating fewer parameters per token, which Alibaba says lowers inference costs. The launch also reflects the company's effort to balance safety with open-source AI releases as concerns over AI-enabled cyberattacks grow.

Qwen4-Max has 2.4 trillion total parameters and activates 71 billion per token, 25% fewer than its predecessor. Alibaba's internal benchmark shows a 60% reduction in average cost per task, but independent testing of these results has not been reported.

The players

Alibaba Group Holding Ltd.

Alibaba Group Holding Ltd. is a Hangzhou-headquartered Chinese technology conglomerate whose e-commerce, cloud computing and AI businesses are among the largest in Asia.

The details

Alibaba introduced cybersecurity guardrails and dedicated datasets with post-training enhancements aimed at mitigating chemical, biological, radiological and nuclear risks. QwenCloud offers the model at 7.5 U.S. cents per 1 million tokens, and Alibaba said smaller-scale, open-weight Qwen4 versions will follow for global developers.

Timeline

  1. Late Sunday, November 15, 2026: Alibaba shares fell more than 1% in the overnight session.

  2. 2026: BABA stock has declined more than 39% year to date.

Roadmap

Alibaba is pairing a closed flagship with a promised open-weight Qwen4 line, continuing the split-release strategy of its Qwen series. The move positions the company in the competitive race among Chinese and U.S. AI developers to lower inference costs while keeping open-source access.

Developers and businesses using QwenCloud can access Qwen4-Max at 7.5 U.S. cents per 1 million tokens, which could lower the cost of building AI-powered tools. Users may also benefit from the smaller open-weight Qwen4 versions Alibaba plans to release for global developers.

The takeaway

Lower per-task costs and open-weight releases could make advanced AI more accessible to developers worldwide, though the efficiency claims remain company-reported. Watch for independent benchmarks as the smaller Qwen4 models arrive.

Further reading

Explore more Artificial Intelligence coverage for context on how rival models are reshaping the field.