Qwen3.5-Max-Preview Debuts With Strong LMArena Rankings
Alibaba launched the preview version of its latest flagship AI model, Qwen3.5-Max-Preview, on March 19. Right away, the model posted strong results on LMArena, the benchmark platform operated by the research group LMSYS that ranks AI chatbots through anonymous human preference voting.
Across several categories, Qwen3.5-Max-Preview landed among the stronger performers. It ranked fifth globally in mathematics with a score of 1,491. On the expert-level leaderboard, it reached tenth place with a score of 1,498.
Chinese-Language Leaderboard Performance
On the Chinese-language leaderboard, Qwen3.5-Max-Preview placed eighth overall. But here's the part that really stands out: it was the highest-performing model developed in China.
That put it ahead of several domestic rivals, including ByteDance’s Doubao 2.0, Zhipu AI’s GLM-5, and Moonshot’s Kimi K2.5.
Alibaba’s Position in the Global AI Model Race
The model’s performance shows Alibaba is competitive in a crowded field, even if it hasn’t taken the top global position.
The top spots on LMArena’s overall leaderboard are still held by Western AI labs. Anthropic’s Claude Opus 4.6 holds the top two positions, followed by Google’s Gemini 3.1 Pro Preview, xAI’s Grok 4.20, and Google’s Gemini 3 Pro.
Competitive, But Not Yet Leading Globally
Those rankings place Qwen3.5-Max-Preview in a strong global position, though not at the very top. It’s performing well enough to stand out, especially in math, expert-level evaluations, and Chinese-language performance, but the overall leaderboard still favors Western developers.
Qwen3.5-Max-Preview Outperforms Major Models in Chinese-Language Benchmarks
According to Futunn, Qwen3.5-Max-Preview outperformed GPT-5.4, Claude 4.5, and Grok 4.1 on the Chinese-language benchmark.
Alibaba also ranked among the top five AI companies globally in LMArena’s company-level rankings.
More Chinese AI Companies Reach the Global Top Ten
AIBase reported that five Chinese companies now appear among the global top ten. Those companies include Alibaba, ByteDance, Zhipu AI, Moonshot, and Baidu.
Qwen3.5 Series Builds on Alibaba’s Fast Release Pace
Qwen3.5-Max-Preview is the latest step in the Qwen3.5 series, which Alibaba first introduced in February 2026. The series was launched with a focus on agentic AI capabilities.
According to Reuters, the Qwen3.5 family is 60 percent cheaper to operate and eight times more efficient at handling large workloads than the previous generation.
Qwen3.5 Model Range and Architecture
The Qwen3.5 series includes eight models, ranging from 0.8 billion to 397 billion parameters.
These models are built on a sparse Mixture-of-Experts architecture. That setup activates only 17 billion parameters per forward pass.
Preview Availability and Ongoing Iteration
Alibaba Cloud said Qwen3.5-Max-Preview is currently available in preview form. The company also said it plans to keep improving the model based on developer feedback.
Following the Earlier Qwen3-Max Release
Qwen3.5-Max-Preview follows Qwen3-Max, which launched in September 2025. At launch, Qwen3-Max ranked third on LMArena’s text leaderboard and had more than one trillion parameters.