Qwen3.7-Max-Preview Enters Arena at Rank 13 — Alibaba Reaches 6th Lab Globally
Alibaba’s Qwen3.7-Max-Preview entered the Arena text and Vision leaderboards on May 14. Results published May 19 show it at rank 13 globally on the text leaderboard, slotting between GPT-5.5 and Grok 4.2. It is the top-ranked Chinese model on the platform.
The entry lifts Alibaba to rank 6 among labs globally on Arena — the first time it has held that position.
Sub-Leaderboard Breakdown
The aggregated rank obscures where the model is genuinely competitive:
| Sub-leaderboard | Rank |
|---|---|
| Mathematics | 7 |
| Expert prompts | 9 |
| Software and IT | 9 |
| Coding | 10 |
| Text (overall) | 13 |
Top-7 in math and sub-10 in expert tasks puts Qwen3.7-Max at a tier above its previous generation on the benchmarks that matter most for research and engineering workflows.
The Plus-Preview variant, which anchors the vision leaderboard, entered at rank 16, placing Alibaba at rank 5 globally in multimodal evaluation — also the top Chinese lab position.
The Acceleration Story
Qwen3.7’s arrival is as notable for its timing as its ranking. Alibaba has shipped three full model generations since February 2026:
- Qwen3.5 — February 2026, open-weight 397B MoE, Apache 2.0
- Qwen3.6 — April 2026, closed flagship (Max-Preview) ranked 3rd on Artificial Analysis at index score 52; open-weight 27B and 35B variants under Apache 2.0
- Qwen3.7 — May 2026, preview on Arena, ranked 13th globally in text
By comparison, in all of 2025, Alibaba shipped two primary Qwen versions. The shift to monthly preview cadences — where community testing runs in parallel with stable-release preparation — mirrors the pace Anthropic adopted with Claude’s rapid 4.x succession.
Where the Gap Remains
Rank 13 is a genuine frontier entry, but the distance to the top is measurable. GPT-5.5 and Grok 4.2, which Qwen3.7 sandwiches, are themselves a tier below GPT-5.5-high, Claude Opus 4.7, and Gemini 3.1 Pro, which hold the top positions. On SWE-bench Verified, Qwen3.6-27B (77.2%) remains roughly 10 points behind Claude Opus 4.7 (87.6%). Closing that gap on agentic coding tasks is where the next generation wins or loses.
The stable release of Qwen3.7 has not been dated. Based on the Qwen3.6 preview-to-production cycle of roughly six weeks, a June window is plausible.