Alibaba Launches Qwen3.8-Max and Preps Open Weight Release

Alibaba officially launched Qwen3.8-Max on August 3, 2026, after previewing it at the World AI Conference in Shanghai on July 19. It is the largest and most advanced model in the Qwen family to date, and the company confirmed that open weights for the model, along with a smaller version called Qwen3.8-27B, will be published on Hugging Face and ModelScope during the week of August 10, 2026.

A 2.4 Trillion Parameter Giant That Runs Light

According to Alibaba, Qwen3.8-Max has 2.4 trillion parameters in total, but uses a sparse Mixture-of-Experts architecture that activates only about 95 billion parameters per inference. That means despite its massive total size, the computational cost per response stays close to that of much smaller models. The model also supports a context window of up to 1 million tokens.

Open Weights: A Reversal in Alibaba's Strategy

The release marks the first time Alibaba has made open weights available for a model in its top tier Max category, reversing a recent trend at the company of keeping its most powerful models closed. Until now, only smaller Qwen family versions had public weights, while Max stayed restricted to paid API access.

The Backdrop: A Race for Open Weights Among Chinese Giants

Alibaba's announcement comes just days after Moonshot's open weight release of Kimi K3, which became the world's largest open weight model. The sequence of releases underscores an intense race among Chinese AI companies for leadership in open models, an area where companies like DeepSeek, Moonshot and now Alibaba compete for adoption among developers outside China.

Pricing on Par With Closed US Models

Even before releasing the weights, Alibaba had already positioned Qwen3.8-Max's API pricing on par with leading closed US models, according to reporting from Forkast and Yahoo Finance, a bet that pairing competitive performance with available open weights can attract both price sensitive companies and those that prefer running the model on their own infrastructure.

Why It Matters for Brazilian Agencies and SMBs

Open weights for a model the size of Qwen3.8-Max give Brazilian agencies and developers the option to run a frontier model on their own or third party infrastructure, instead of depending solely on the official API, which matters for anyone handling sensitive customer data or trying to cut costs at scale. Even though running a 2.4 trillion parameter model remains out of reach for most SMBs without access to specialized hardware, smaller cloud providers and inference services tend to quickly add this kind of open model to their catalogs, widening the options available in Brazil's AI automation market.