Alibaba previews new Qwen AI models before launch

Alibaba has released preview versions of its upcoming Qwen3.7 AI models ahead of a planned launch at its Alibaba Cloud Summit in Hangzhou on May 20, as it steps up spending on AI.

The preview models ranked among the top AI systems globally on LM Arena.

They outperformed other Chinese AI labs but remained behind models from Anthropic, Google, and OpenAI.

Companies often post preview models on LM Arena, which ranks systems based on user preferences, to collect feedback before final releases.

Alibaba launched its Qwen3.6 series a month earlier and kept its Qwen3.6-Max and Qwen3.6-Plus models proprietary through paid API access.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

Alibaba is spending heavily to grow its AI business

Alibaba’s AI push is expensive. In its latest quarter, adjusted EBITA fell 84%, while free cash flow moved to an outflow of 17.3 billion yuan (US$2.5 billion) 1.
The company is choosing market share over near-term profit. ceo Eddie Wu is pushing “token consumption,” meaning use of AI models, to lift adoption and growth 1.
That bet is bringing in sales. Revenue from AI-related products has grown at a triple-digit rate for 11 straight quarters 2.
Model and application services are also becoming a steadier business, with annualized recurring revenue expected to reach 30 billion yuan (US$4.41 billion) by the end of the year 1.

AI competition now turns on fit, skill, and price

A single leaderboard rank can hide real differences in what a model does well.
On code review benchmarks, which test how AI models analyze software, Claude from Anthropic stands out in deep logic analysis. Qwen performs well across tasks and pulls different views into practical fixes 3.
Businesses may end up using teams of specialized AI agents that cover one another’s weak spots 3.
Alibaba is leaning into that approach with models beyond its flagship, including the lower-cost Qwen-Flash series. That lets developers match each task with a cheaper or stronger option 4.