Top 5 This Week

spot_img

Related Posts

Alibaba Unveils Qwen4-Max AI Model With 1M-Token Context

Alibaba Unveils Qwen4-Max AI Model With 1M-Token Context

Alibaba has unveiled Qwen4-Max, a flagship AI model with built-in image and video understanding, a one-million-token context window and tools for projects that can run for hours. The company showed the model building a playable game, planning factory production lines and carrying out research.

In Alibaba’s Work360 benchmark, covering 26 domains, 182 subdomains and 721 specialised fields, the average cost per task fell from $4.43 to $1.67 against Qwen3.8-Max. The new model scored 52.69, up from 47.74. ProPakistani reported that the model also used fewer tokens, cutting use by 24 and 35 per cent at high and very high reasoning effort on Work360, and by 48 and 49 per cent on Terminal-Bench 4.0.

Inside Qwen4-Max

Qwen4-Max runs on a 2.4-trillion-parameter backbone that activates 71 billion parameters per token, with another 328 billion for recognising nearby token patterns. Gated DeltaNet and sparse attention help it process long inputs.

Alibaba reported stable training, with 25 per cent fewer active parameters than the previous base model, though two competing checkpoints got extra training.

Games, video and factories

Alibaba showed the model building and rendering 3D scenes in Blender for games and 3D printing. It supports Unity, Unreal Engine and Godot, and built Meow of Magic, a playable Unreal Engine game. With WonderClip, it took a video from storyboard through rendering and editing.

At SANF and Pacific Union, it turned 2D layouts into 3D car production lines, simulated cycle times, tested controllers and designed parts. In one demo, manual programming time for a part fell from 475 minutes to 24.

Finance, law and healthcare

In finance, Qwen4 used 90 tools to gather market data, cross-check disclosures and analyse CSI 300 derivatives, producing an Excel workbook and a matching PowerPoint deck.

In law, it sorted 198 pre-IPO papers into 14 folders and produced an 86-page report covering 12 review areas and 14 legal issues, with citations.

In healthcare, it reviewed six days of records, including 1,433 medical images, in a critical-care demo. Alibaba said its imaging finding matched the attending doctor’s diagnosis.

Agents that work in stages

The open-source Live Delivery plugin lets agents hand over an early version of their work, then revise it as users give feedback. In cloud scheduling tests, Alibaba reported lower model-call costs and a more than 14-fold speedup in solving an offline CPU scheduling problem.

In a 48-hour run on Deep Principle’s Mira platform, Qwen4 screened 2.7 million molecules and produced a chip coolant sample. In an Alzheimer’s demo, it ran 3,239 evaluations and developed 24 candidates, one with a binding affinity 3.6 times the reference compound. Lab validation is still needed.

Safety and pricing

In secure-coding tests, functions passing both functionality and security checks rose from 22 per cent to 31.8 per cent. Alibaba also reported better handling of risky chemical, biological, radiological and nuclear requests.

Qwen4-Max is available through the QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder, with web search, tool calling and structured outputs. It costs $1.50 per million input tokens, $4.50 per million output tokens and $0.075 per million cached input tokens, with the cache rate down 70 per cent. Alibaba plans to release the weights of smaller Qwen4 models later.

What it means for developers

Qwen4-Max points to a shift in how AI models are used. Earlier models answered one question at a time. Agentic models take on multi-hour jobs, use tools, check their own work and hand over results in stages, so the unit of work for software teams becomes a whole task, not a single prompt.

The same shift is reshaping the AI industry. Nvidia’s leadership is under scrutiny as Jensen Huang weighs a comeback after his exit, and new models keep arriving, including Mistral’s Large 4 with one trillion parameters. In Pakistan, entry-level developer jobs fell 20 per cent as AI took over routine coding, while Jazz’s ROX AI hub brings Urdu and English voice assistants to local users.

The question now is not whether AI can write code, but whether it can finish a job, check it and fix it. Qwen4-Max is Alibaba’s answer.

Frequently asked questions

What is Qwen4-Max?

It is Alibaba’s flagship AI model, with image and video understanding, a one-million-token context window and tools for long projects such as building games and planning factories.

How much does Qwen4-Max cost to use?

Through the QwenCloud API: $1.50 per million input tokens, $4.50 per million output tokens and $0.075 per million cached input tokens.

What can Qwen4-Max build?

Alibaba showed it building a playable Unreal Engine game, rendering 3D scenes in Blender and turning 2D layouts into 3D car production lines. It also handled finance, law and healthcare tasks.

Will smaller Qwen4 models be released?

Yes. Alibaba plans to release the weights of smaller Qwen4 models later.

Where can developers use Qwen4-Max?

Through the QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder, with web search, tool calling and structured outputs.

Feature Pakistan News Desk
Feature Pakistan News Desk is the editorial team of Feature Pakistan, an independent digital news platform based in Pakistan. The desk reports national, world, business, sports, technology, health and education news, and every story is cross-checked against multiple sources before it is published. To reach the newsroom — including for corrections — email featurepakistanofficial@gmail.com or use the Contact page at featurepakistan.com/contact-us/.

opinion