Alibaba Group is stepping up its artificial intelligence game in a big way. The company has just introduced Qwen-3-Max-Preview, its largest AI model so far, joining the likes of OpenAI, Google DeepMind, and other tech giants building trillion-parameter systems.
The model, which has over 1 trillion parameters, went live on Alibaba Cloud’s platform as well as the AI model marketplace OpenRouter on Friday.
From Millions to Trillions
Qwen-3-Max-Preview is the latest in Alibaba’s Qwen3 series, which first launched in May. Earlier versions in the lineup ranged from 600 million to 235 billion parameters. Now, Alibaba has broken into trillion-scale territory, marking a huge leap forward for its AI ambitions.
For those unfamiliar, parameters are the “knobs and switches” that shape an AI model’s intelligence. The more parameters a model has, the more advanced it tends to be – though it also takes much more computing power to train and run.
How It Stacks Up Against Rivals
Alibaba says this new model outperformed its earlier top performer, the Qwen3-235B-A22B-2507, in internal tests. The company also shared results suggesting that Qwen-3-Max-Preview beat out MoonShot AI’s Kimi K2, a non-reasoning version of Anthropic’s Claude Opus 4, and DeepSeek V3.1 across five benchmarks. These results weren’t part of a formal technical report, though.
“Qwen3-Max-Preview shows substantial gains … in overall capability, with significant enhancements in Chinese-English text understanding, complex instruction following, handling of subjective open-ended tasks, multilingual ability, and tool invocation,” Alibaba said. “Scaling works – and the official release will surprise you even more.”
For context, OpenAI’s GPT-4.5 is believed to have somewhere between 5 and 7 trillion parameters, keeping it among the largest in the world.
Open Source Leader, But Not This Time
Alibaba’s Qwen series has been a massive hit in the open-source community, with more than 20 million downloads and 100,000 spin-off models on Hugging Face. That success has put the company at the top of the global open-source AI ecosystem.
However, Qwen-3-Max-Preview isn’t being open-sourced just yet. Access is currently limited to official channels, much like its predecessor, Qwen2.5-Max, which launched back in January.
One of Alibaba’s engineers, Binyuan Hui, teased on social media that a “thinking” version of the model is “on the way.”
Pricing and Competition
On Alibaba Cloud, using Qwen-3-Max-Preview comes with premium pricing. It starts at US$0.861 per million input tokens and US$3.441 per million output tokens, making it one of the priciest Qwen models available through API.
By comparison, the older Qwen3-235B-A22B-2507 costs US$0.287 per million input tokens and US$1.147 per million output tokens for its non-thinking version. Meanwhile, Kimi K2’s newest model is priced at US$0.60 per million input tokens and US$2.50 per million output tokens.
A Massive AI Investment
Alibaba is backing these advancements with serious money. The company has pledged 380 billion yuan (US$52 billion) toward AI infrastructure over the next three years – a bigger commitment than what it has spent in the past decade combined.
The payoff is already showing. According to Alibaba’s latest financial results, its AI-related products have delivered triple-digit growth for eight straight quarters.
ALSO READ: Apple Event 2025: iPhone 17 Pro and Everything Else to Expect
