Alibaba launches flagship Qwen4-Max model without open weights
The 2.4-trillion-parameter Qwen4-Max accepts text, images and video with up to one million tokens of context, and Alibaba says it cuts average task cost by about 62%, but unlike many earlier Qwen models its weights cannot be downloaded.

Alibaba on November 16 launched Qwen4-Max, the new flagship of its Qwen model family, pitching it as an AI teammate for work, creativity and discovery. Unlike many earlier Qwen releases, the company is keeping the model's weights closed: it is available through the QwenCloud API and Alibaba's apps but cannot be downloaded or modified. Alibaba said smaller Qwen4 models will get open weights, without giving a date or license.
Specs and pricing
Qwen4-Max is a 2.4-trillion-parameter mixture-of-experts model that reads text, images and video and accepts up to one million tokens in a single prompt, TNW reported; VentureBeat noted that 71 billion parameters are activated per token. On Alibaba's internal Work360 evaluation, the average cost of a task fell from $4.43 to $1.67, roughly 62%, compared with Qwen3.8-Max.
The QwenCloud API charges $1.50 per million input tokens and $4.50 per million output tokens, undercutting several prominent models from Anthropic, OpenAI and Google. The model is also available in Qwen Studio, QwenWork and Qoder.
A mixed benchmark picture
In Alibaba's own chart, Qwen4-Max scored 75.7 on the DeepSWE v1.1 coding test, just ahead of Claude Opus 5.5 at 74.2, but 48.5 on Terminal-Bench 4.0, behind Opus 5.5 (59.6) and Kimi K3.1 (67.2). On Artificial Analysis's aggregate Intelligence Index it scores 43, behind leading US models and Moonshot AI's Kimi K3.1 Pro at 46, VentureBeat reported.
Alibaba also highlighted safety work: it kept training data for defensive cyber tasks such as finding and fixing software flaws but removed data aimed at boosting attack skills, and added training to reduce chemical, biological, radiological and nuclear risks.
Sources: TNW, VentureBeat, Alibaba Cloud
Get the latest news delivered to your inbox.
Related articles




