around.pk
Monday, 16 November 2026
☀ Karachi 31°☀ Lahore 25°☀ Islamabad 22°
Technology

Alibaba Launches Qwen4-Max AI Model Capable of Building Games and Designing Car Parts

Alibaba's new flagship AI model, Qwen4-Max, can create playable games, design automotive parts and automate complex industrial workflows, with the company reporting sharp gains in efficiency and lower operating costs.

around.pk·3m ago

Alibaba has introduced Qwen4-Max, a flagship artificial intelligence model that combines image and video understanding with a context window capable of handling one million tokens, allowing it to work on complex projects that can run for hours at a stretch. The company demonstrated the model building a playable game, planning factory production lines and conducting research tasks.

In Alibaba's internal Work360 benchmark, which spans 26 broad domains, 182 subdomains and 721 specialised fields, Qwen4-Max scored 52.69, up from 47.74 for its predecessor. The average cost per task dropped from $4.43 to $1.67, while the model used between 24% and 35% fewer tokens at high reasoning settings and nearly half the tokens on the Terminal-Bench 4.0 test.

The model is built on a 2.4-trillion-parameter backbone that activates 71 billion parameters per token, with an additional 328 billion dedicated to recognising nearby token patterns. Alibaba said techniques such as Gated DeltaNet and sparse attention help it process long inputs, while four gated residual branches manage how information moves between layers. The company reported stable training and 5% to 13% lower bits per byte across its knowledge, code, question-answering and STEM evaluations.

On the creative and industrial side, Alibaba showed Qwen4-Max building and rendering 3D scenes in Blender for games and 3D printing, with support for Unity, Unreal Engine and Godot. It produced a playable Unreal Engine game called Meow of Magic and can refine gameplay and visuals using feedback from test runs. At automotive operations SANF and Pacific Union, the model helped convert 2D layouts into 3D production lines, simulate cycle times, test controllers and design parts. In one demonstration, manual programming time for a part fell from 475 minutes to 24 minutes.

Alibaba also highlighted scientific applications. In a 48-hour run on Deep Principle's Mira platform, Qwen4 screened 2.7 million molecules and produced a coolant sample for high-power chips after four rounds of design and testing. In a separate Alzheimer's research demonstration on PharmMaster, it made 454 tool calls and developed 24 candidate molecules, though the results still require laboratory validation.

The model is available through the QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder. API pricing is set at $1.50 per million input tokens, $4.50 per million output tokens and $0.075 per million cached input tokens. Alibaba said it plans to release the weights of smaller Qwen4 models at a later date.

The story so far Alibaba

  1. Nvidia's Jensen Huang Lobbied Trump on China Chip Sales, Report Says
  2. Experts Say Nvidia CEO's Alleged Affair Could Have Exposed Sensitive Company Details
See the full timeline →
Artificial intelligenceChinaAlibabaBlenderQwen4-MaxQwenCloud APIUnreal Engine

Continue the story

View all →

Read more