Alibaba has introduced Qwen 4 Max Preview, a flagship multimodal reasoning model built for programming, software interface creation, and complex logic tasks. The system features a large 1 million token context window capable of ingesting text, code, images, and video files. The preview is now accessible through Qwen Chat and cloud accounts, though early community tests show mixed results on visual tasks.
Alibaba engineered the model to combine logical deduction with multimodal processing across massive datasets. The company has not revealed the parameter size of the network or indicated if the weights will become open source. Official benchmark scores also remain unpublished, forcing early users to gauge performance through manual testing.
Early developer reports suggest strong capabilities in code generation, with some users matching its programming speed and accuracy to DeepSeek V4.1 Flash. Visual reasoning presents clear hurdles. Early testers report that the system fails to handle basic animation logic, struggling to depict spinning bicycle wheels or acceleration motion lines around vehicles. Another user reported spending 1 full hour generating simple voxel artwork that produced mediocre visual quality.
Users can run the preview through the Qwen Chat interface or deploy workloads via Qwen Cloud. Cloud deployments require a paid Token Plan. The Lite tier costs $6 monthly for 20k credits, the Essential tier costs $10 for 45k credits, the Standard tier costs $18 for 80k credits, and the Pro plan costs $43 for 200k credits.
