Qwen 3.8 Max Preview: The New Chinese AI Model
A new Chinese AI model was released just yesterday. It's called Qwen 3.8 Max Preview, and it's currently in preview—meaning it's free to use with 1 million credits available at no cost. Among Chinese AI models, the ones that are genuinely powerful can be counted on one hand. Qwen has consistently been one of the strongest, particularly for image editing and code generation. This new preview version pushes those capabilities further with a 1-million-token context window, an updated Mixture of Experts architecture for faster inference, and built-in reasoning that's activated by default. Here's a hands-on look at what this model can do, tested on real tasks.
Interactive Arabic Presentation for Children
The first test was a practical one: create an interactive, animated presentation in Arabic for children about keeping the ocean clean. The prompt requested animated elements—swimming fish, rising bubbles, moving waves, a rotating sun—plus an interactive game where children click on trash to remove it from the ocean, a quiz with instant feedback, and a pledge button. The presentation needed to work on both mobile and desktop.
The model generated over 1,000 lines of code in seconds. The speed is noticeable—this is running on Qwen's own cloud infrastructure during the preview period, so there's no cost to the user and no throttling. The output was a complete HTML file with eight interactive slides.
The presentation opened with the title "Keep the Ocean Clean" in Arabic, set in a carefully chosen font. The first slides contained educational content: why the ocean matters, how it produces more than half the oxygen we breathe, and what threatens it—plastics, glass, oils, and waste. Each slide had simple but effective animations. The icons and emojis were drawn from standard libraries, but they were used appropriately and looked clean.
The interactive game appeared as promised. Clicking on trash items removed them from the ocean scene with visual feedback. The quiz section asked questions about ocean conservation and provided instant feedback—correct answers triggered colored confetti animations, while incorrect answers prompted another attempt. The logic worked correctly. The final slide presented a pledge: "I promise to protect the ocean and its creatures. I will not throw trash." A button allowed the child to accept the pledge.
The code ran with zero errors. For a preview model generating over 1,000 lines of complex interactive code, this level of reliability is exceptional. This is what makes modern AI models genuinely threatening to traditional programming workflows—not just that they can write code, but that they can write it correctly, explain how to run it, and deliver a complete working product in seconds. The days of paying a developer to build simple interactive widgets are effectively over.
Python Coding Test
The second test switched to Python. The goal was to evaluate pure coding capability in a different language. The model was fast—noticeably fast. The 2-million-token context window and the updated Mixture of Experts architecture mean that not all expert modules fire at once; a gating mechanism selectively activates only the relevant experts for each query. This is what makes the model both powerful and efficient.
The Python output was clean, well-structured, and error-free. Minor stylistic preferences might vary, but there were no logical errors, no syntax mistakes, and no hallucinations in the code. The model handled the task with the same reliability it showed in the HTML presentation test.
Arabic Language Support
Arabic language support has been a consistent strength of the Qwen series relative to other Chinese models, and this preview maintains that capability. The Arabic text in the generated presentation was grammatically correct, contextually appropriate, and natural-sounding. It doesn't match the fluency of models like Gemini, which remains the strongest Arabic-language model currently available, but it performs at a medium-to-good level—usable for real projects without requiring extensive correction.
This matters because very few Chinese AI models handle Arabic well at all. Most produce garbled text, incorrect grammar, or refuse to engage with Arabic prompts entirely. Qwen's Arabic capability makes it a practical option for Arabic-speaking developers and content creators who want a free, fast, capable model.
What the Model Offers
The Qwen platform provides multiple models beyond the 3.8 Max Preview. These include image generation models, video generation models, image editing capabilities, and video editing tools. The text models are the core offering, and the 3.8 Max Preview represents the latest iteration. During the preview period, 1 million credits are available for free, making this an accessible way to test the model's capabilities without commitment.
The comparison feature that allows side-by-side evaluation of different model versions is currently unavailable but expected to be added in future updates.
A Note on Version Switching
During testing, the model briefly switched from 3.8 back to the more stable 3.7 version. This appears to be a preview-period behavior where the system occasionally reverts to the stable release. Both versions share the same credit pool. The important point is that the 3.8 Max Preview exists, it's accessible, it's powerful, and it's free to use right now.
