Alibaba Qwen 3.7 Max ranks #13; Plus to be open source

Alibaba posted Qwen 3.7 Max-Preview and Qwen 3.7-Plus-Preview to Arena AI on May 14. The Max preview placed #13 in Arena’s Text leaderboard. The Arena ranking places Alibaba sixth among AI labs in text and fifth in vision. Qwen 3.7 Plus will be released as open source while Qwen 3.7 Max will remain proprietary ahead of the Alibaba Cloud Summit on May 20.

Both models appeared on Arena AI five days before the Cloud Summit. Arena uses blind, crowd-sourced comparisons based on user preferences to rank models. Both Qwen 3.7 variants are listed in preview mode with web search and the code interpreter disabled.

On Arena, Qwen 3.7 Max ranked #13 overall in Text Arena, #7 in math, #9 on expert-level prompts and #9 in software and IT tasks. An official Qwen blog post states that the Plus variant will be open source and that Max will be offered on a paid tier. Alibaba plans to publish official API pricing and a full release timetable after the Cloud Summit.

In a technical test of the preview build, Qwen 3.7 Max solved a demanding algebraic task involving a degree-19 Dickson polynomial. The model factored components over the complex numbers, set up a recurrence to compute a large integer result and ran cross-checks using modular arithmetic against multiple moduli. The numerical answer matched the model’s internal verification steps.

In coding tests the model produced concise, readable code and showed explicit design choices. For a game-design prompt Qwen 3.7 Max produced a 2D implementation with named enemy roles, hideout zones and detection logic. When asked, the model extended that design toward a 3D aesthetic. The preview output used fewer lines of code while preserving functionality.

In creative writing tests the model generated compact, setting-driven fiction. In one mystery prompt the model produced a plausible case but did not reconcile timeline details; the unresolved timeline led to an incorrect conclusion when the solution depended on temporal checks.

The preview restrictions disable autonomous multi-step runs, live web access and the code interpreter, so those features were not assessed. Final capabilities, feature sets and pricing are expected after Alibaba’s Cloud Summit on May 20.

Alibaba’s Qwen account posted on social media: “Qwen3.7 Preview lands on Arena! Here come Qwen3.7-Max-Preview & Qwen3.7-Plus-Preview. Alibaba now #6 lab in Text, #5 in Vision. Can’t wait to release Qwen3.7 series models! Stay tuned!”