01What Is Qwen 3.8 Max and How Does It Work?
Qwen 3.8 Max is Alibaba's latest AI model that can work alone for days. You give it a goal, and it plans, codes, tests, and fixes its own work without you watching. It is free to use right now in your browser, and the open weights come out next week.
The model uses a loop: it turns feedback into tasks, does the tasks, tests the results, and merges finished work. In a public test, it ran alone for 16 days and handled 2 commits, 127 pull requests, and 151 issues on GitHub. That is like a worker, not a chatbot.
It also beat 458 human teams in a 24-hour competition. The human just describes the goal in plain words; the AI does the rest.
02Main Features of Qwen 3.8 Max
- 2.4 trillion parameters with 95 billion active at any moment.
- 1 million token context window for long documents, videos, and chats.
- Open weights next week, plus a smaller 27B version.
- Free chat at chat.quen.ai without login.
- Generates images and videos, something Claude cannot do.
- Runs your computer like clicking, typing, and using apps.
- Works with your tools like Claude Code, Codex, and OpenCore.
It is the first Max-class model Alibaba has open-sourced. No other lab has released weights at this scale.
This model was built to be a walk-away engine. You can set a goal, leave, and come back to finished work.
03How to Get Started with Qwen 3.8 Max
- Go to chat.quen.ai in your browser. You do not need to log in.
- Type a real task from your business, like building a landing page or making images.
- If you want to run it locally, wait for the weights next week.
- Download the open weights and use a tool like LM Studio or Ollama to run it.
- Plug it into your existing agent setup using the API or OpenRouter.
- For a full system, join the Air Profit Boardroom to get the Agent OS with Qwen connected.
You can also test it on your own tasks before the weights drop. Give it a coding project and watch it work.
04How to Use Qwen 3.8 Max for Your Work
For coding, give Qwen a project idea and a folder. It will create code, test it, and fix bugs. In one test, it rebuilt an experiment from a paper and beat it by 2.7 points on a math benchmark.
For computer use, Qwen can operate your computer like a human. It can open apps, click buttons, and type text. This makes it useful for automating tasks like data entry or research.
For vision tasks, it reads documents, charts, and screenshots accurately. It can watch over 100 hours of video and remember everything.
To get the best results, run multiple chats in parallel. One for a website, one for images, one for an app. That way, you get more output in the same time.
Give feedback when the output is not right. Qwen usually needs two or three rounds of fixes to get to where you want.
05Benchmarks: How Qwen 3.8 Max Compares
On the LMArena leaderboard, Qwen 3.8 Max is number two in the world with a score of 1496. Claude Fable 5 is number one with 1509. Meta is third, Google is fourth, and Moonshot is fifth.
OpenAI is sixth.
On OS World Verified, which tests real computer use, Qwen scored 86.1. That beats Claude Opus 4.8 at 83.4 and Claude Fable 5 at 85.
On PaperBench, reproducing research, Qwen scored 93. That is ahead of GPT-5.6 Sol at 90.5.
In our own tests on GoldieBench, Qwen landed in the top 5 of every model we have tested. It built some games better than G55.6 Sol.
While Claude still wins most coding tests, Qwen wins on vision and computer use. It is the first open model to challenge the top closed US models.
06Tips, Limits, and Mistakes to Avoid
Avoid the deep research feature. In my tests, it fabricated sources and cited an Instagram post. Use other tools for research and stick to building with Qwen.
Expect slow responses during peak hours, especially on the free plan. Fix this by running multiple chats in parallel instead of waiting for one.
Qwen needs more feedback than Claude. Do not expect it to nail everything first try. Test the output and tell it what is broken.
If you want to run it locally, you need powerful hardware like an RTX 5090 or a DGX Spark. A Mac Studio may not handle it well.
Do not treat Qwen as a replacement for everything. Use it for building, not for research. And always verify important outputs.
07Who Should Use Qwen 3.8 Max and Why It Matters
Qwen 3.8 Max is for anyone who wants free frontier AI. Small businesses can use it to build websites, create images, and automate tasks without paying for expensive APIs.
Non-technical people can use it well because the AI does the coding, testing, and fixing. The only skill you need is describing your goal clearly.
This model matters because it proves open-source AI can compete with closed US models. It also brings down the cost of intelligence for everyone.
Use it to stay ahead. The gap between people who use agents and those who do not is growing. Start using Qwen now to see what it can do for you.
08The Walk-Away Engine: A New Way to Work
Old AI needed you to type, check, and type again. Qwen 3.8 Max lets you set a goal and walk away. It runs a loop: plan, build, test, fix, and repeat.
Alibaba proved this in three tests. It built a coding tool alone for 16 days. It improved a research paper alone.
It beat 458 human teams in a 24-hour contest.
Your job is to give clear instructions. The model handles everything else. This changes how you work, letting you finish projects while you sleep.
Start small. Give Qwen a simple task today and watch how it works. Then scale to bigger projects.
