Qwen 3.7 Max Preview Arrives! Two Generations of Large Models Iterate in Parallel, Accelerating Despite Lin Junyang's Departure

Alibaba's Qwen team has once again surprised the AI community with the sudden release of the Qwen 3.7 Preview series. The official Arena account (formerly the Large Model Arena) announced the performance of Qwen3.7-Max-Preview and Qwen3.7-Plus-Preview, showing strong results across text and vision benchmarks.

In the text domain, Qwen3.7-Max-Preview achieved an overall rank of 13th on the Arena leaderboard, lifting Alibaba's (lab) ranking to 6th. It is also the highest-ranked Chinese model on the list. In the vision domain, Qwen3.7-Plus-Preview secured the 18th spot, propelling Alibaba to 5th place overall—again the top domestic model.

Beyond overall rankings, the model demonstrated impressive performance in specialized areas. Qwen3.7-Max-Preview entered the global top 10 in mathematics (7th), expert prompts (9th), software/IT (9th), and coding (10th). In the Expert Arena, it ranked 9th when limited to expert prompts, with Xiaomi's Mimo v2.5 Pro taking 7th.

The release marks a clear acceleration in Qwen's iteration cycle. Since the Qwen3 series, major version updates have shrunk from 4-6 months to just 2-3 months, and in 2026, new preview versions have appeared almost monthly. This rapid cadence reflects a shift toward frequent experimentation and high-frequency delivery.

Notably, the team has maintained momentum even after the departure of Qwen's former leader Lin Junyang, who left during the Qwen3.5 era. His parting message—"Qwen brothers, just keep doing what you were doing"—seems to have set a steady course. The subsequent Qwen3.5, 3.6, and now 3.7 versions continue to improve in speed and capability.

As the community eagerly awaits the official release and open-source availability, the Qwen team's fast-paced innovation shows no signs of slowing down.