Alibaba Qwen3.7 previews rank as top Chinese AI models on LM Arena
Synopsis
Alibaba's Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview have leapt to the top of LM Arena's Chinese AI rankings — 13th and 16th globally — just weeks after the Qwen3.6 launch, with a full reveal expected at the Alibaba Cloud Summit in Hangzhou on 20 May 2026.
Key Takeaways
Alibaba Group Holding released preview versions of its Qwen3.7-Max and Qwen3.7-Plus models on 19 May 2026 .
The previews rank 13th globally in text and 16th globally in vision on LM Arena 's leaderboard, making them the highest-ranked Chinese AI models.
Both previews still trail Anthropic Claude , Google Gemini , and OpenAI GPT models in overall rankings.
The Qwen3.7 series follows the Qwen3.6 launch by approximately one month, reflecting an accelerating development cycle.
Alibaba is expected to formally announce the full Qwen3.7 series at the Alibaba Cloud Summit in Hangzhou on 20 May 2026 .
Like its predecessor, Qwen3.7 is expected to be proprietary and API-gated, continuing Alibaba 's shift away from full open-sourcing of frontier models.
Alibaba Group Holding has unveiled preview versions of its next flagship AI model series, with Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview claiming the top spots among Chinese AI models on LM Arena's global leaderboard as of Tuesday, 19 May 2026. The previews rank 13th globally in text capabilities and 16th globally in vision capabilities, according to the benchmark firm.
Where the models stand globally
While the Qwen3.7 previews lead all Chinese AI labs across several evaluation categories on LM Arena — which ranks models based on aggregated user preferences — they still trail leading US products. Anthropic's Claude, Google's Gemini, and OpenAI's GPT models all rank ahead of the new Alibaba previews on the leaderboard.
The competitive backdrop
The Qwen3.7 series arrives roughly one month after Alibaba launched its previous flagship Qwen3.6 series, signalling an accelerating release cadence. Industry observers note that tech companies routinely publish preview models on LM Arena to gather user-preference data and refine the final product before a full launch.
Alibaba has not disclosed technical specifications or a formal release date for the Qwen3.7 series, but has teased a major model announcement at the Alibaba Cloud Summit in Hangzhou, scheduled for Wednesday, 20 May 2026.
Why it matters: the open-source pivot
Alibaba is among a growing cohort of Chinese AI developers that have recently moved away from fully open-sourcing their most powerful models, as commercialisation pressures mount. The company kept its earlier Qwen3.6-Max and Qwen3.6-Plus models proprietary, available only through a paid API (application programming interface). The shift reflects the enormous compute costs associated with frontier model development, according to industry analysts.
What's next
All eyes are on the Alibaba Cloud Summit in Hangzhou on 20 May 2026, where the company is expected to formally unveil the full Qwen3.7 series. A competitive launch could intensify pressure on rival Chinese AI labs — and force a fresh reckoning for global incumbents — as Alibaba continues to scale its AI infrastructure spending.
Point of View
A pattern that mirrors how OpenAI and Google have used benchmark positioning as a marketing instrument. What mainstream coverage underplays is the open-source retreat: Alibaba and its Chinese peers are quietly closing the open-weight tap as API monetisation becomes existential, which narrows the cost advantage that made Chinese models disruptive in the first place. The gap to Claude, Gemini, and GPT at the frontier remains real, and closing it demands compute spend that increasingly justifies proprietary lock-in. Enterprises evaluating Chinese AI alternatives should watch whether the Qwen3.7 full release matches the preview's Arena trajectory — or whether the benchmark lead evaporates on production workloads.
NationPress
6 Jul 2026
Frequently Asked Questions
What are the Alibaba Qwen3.7 preview models?
The Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview are pre-release versions of Alibaba 's next flagship AI model series, published on LM Arena on 19 May 2026 . They rank 13th globally in text and 16th globally in vision capabilities, topping all other Chinese AI models on the leaderboard.
How do Alibaba's Qwen3.7 models compare to ChatGPT and Claude?
Qwen3.7 previews lead Chinese competitors but still rank below OpenAI 's GPT , Anthropic 's Claude , and Google 's Gemini on LM Arena 's global leaderboard as of 19 May 2026 . The gap to top US models remains the key benchmark challenge for Alibaba .
When will Alibaba officially launch the Qwen3.7 series?
Alibaba has teased a major model announcement at the Alibaba Cloud Summit in Hangzhou on Wednesday, 20 May 2026 . The company has not yet disclosed full technical specifications or pricing for the Qwen3.7 series.
Why is Alibaba moving away from open-sourcing its AI models?
Alibaba and other Chinese AI developers are increasingly keeping frontier models proprietary due to rising commercialisation pressures and the massive computing costs required to build them. The Qwen3.6-Max and Qwen3.6-Plus were already restricted to paid API access, and the Qwen3.7 series is expected to follow the same model.
What is LM Arena and how does it rank AI models?
LM Arena is a benchmark platform that ranks AI models based on aggregated user preferences from head-to-head comparisons. Tech companies often release preview models on the platform to collect real-world feedback and optimise their final product before a public launch.
