On August 3, 2026, Alibaba released Qwen3.8-Max — the most powerful model in its Qwen lineup to date. We break down its specifications, benchmarks and why this is another step in the fast-moving race of large Chinese AI models.
What Is Qwen3.8-Max
Qwen3.8-Max is Alibaba Cloud’s flagship model with 2.4 trillion parameters, slightly smaller than Kimi K3 from Moonshot AI (2.8 trillion), which still remains the largest open model in the world. The model’s context window is up to 1 million tokens, allowing it to process huge amounts of text, images and video in a single request.
Architecture: Efficiency Through MoE
The model is built on a sparse Mixture-of-Experts architecture — instead of activating all 2.4 trillion parameters at once, only about 95 billion are used for each request. This significantly reduces compute cost and response latency, allowing the model to process text, images and video without a proportional increase in expenses.
Capabilities and Benchmarks
According to Alibaba, Qwen3.8-Max delivers results comparable to Anthropic’s Claude Fable 5, and in some cases better. The model ranked fifth in Text Arena and second in Vision Arena, trailing only Fable 5 in that category.
Among the stated capabilities are application development, research work, legal document review, sports analytics, financial research and even architectural 3D modeling. Alibaba also claims that the model is capable of autonomously writing code for weeks with minimal human intervention.
Open Weights — Next Week
Alibaba plans to publish the model’s full weights within the week following release — this will be the company’s first “Max”-class model to become fully open. For now, Qwen3.8-Max is available through the API on Alibaba Cloud Model Studio for developers worldwide.
Context: The Race of Chinese Open Models Is Accelerating
The release of Qwen3.8-Max happened less than a month after Moonshot AI released Kimi K3 — a model the company called the world’s first open “3-trillion-class” model. That same week, China’s DeepSeek released the V4-Flash model, which researchers estimate could turn out to be one of the cheapest models in the world to run.
Together, these releases show how quickly Chinese technology companies are increasing the pace of launching powerful open models, competing not only on size but also on deployment cost.
In Brief: The Key Points
- August 3, 2026 — Alibaba released Qwen3.8-Max, with 2.4 trillion parameters
- Context window — up to 1 million tokens
- MoE architecture activates only ~95 billion parameters at a time, reducing cost
- Results comparable to Claude Fable 5, 5th place in Text Arena, 2nd place in Vision Arena
- Full open weights — within the week following release
- Released less than a month after Kimi K3 (2.8T) from Moonshot AI
FAQ
What is Qwen3.8-Max? Alibaba Cloud’s most powerful model in the Qwen lineup to date — 2.4 trillion parameters, with a context window of up to 1 million tokens.
How is Qwen3.8-Max different from Kimi K3? Kimi K3 from Moonshot AI is slightly larger (2.8T parameters versus 2.4T) and still remains the world’s largest open model. Both are Chinese open models released less than a month apart.
Will Qwen3.8-Max be an open model? Yes, Alibaba plans to publish the model’s full weights within a week after release.
How does Qwen3.8-Max compare with Claude Fable 5? According to Alibaba, the model shows comparable and in some cases better results on benchmarks, although in the Vision Arena category it trails Fable 5, taking second place.
What is Mixture-of-Experts architecture? An approach in which the model activates only part of its parameters for each specific request, rather than all of them at once — reducing compute costs and response latency.
The article was prepared by the TechVisor team — practical IT media for people.
