‹ rbtfl

Alibaba's Qwen 3.7 goes API-only at the top, open-weight below

Qwen3.7-Max and the cheaper Qwen3.7-Plus agent variant are proprietary; the open Qwen 3.6 line stays on Apache 2.0

AI· active Whose Money·What They're Not Saying ·5 takes · ·rbtfl upd Jun 25, 2026
post

The split

The same story, as told by newsrooms in different countries. Their words, attributed and linked.

United States

CNBC

“Alibaba unveils Qwen as China's chatbot race shifts to AI agents.”

China tech / strategyread the original ↗

Global

AI/ML API Blog

“Qwen3.6-35B-A3B remains the open-weight top pick, Apache 2.0, free for commercial deployment.”

developerread the original ↗

post

Summary

Alibaba split its Qwen line by tier in mid-2026. Qwen3.7-Max, unveiled at the Apsara Summit on 20 May, is the proprietary flagship; Qwen3.7-Plus, a multimodal agent variant at roughly one-sixth the per-token cost, followed on 1 June. Both are API-only via Alibaba Cloud, not open-weight. The open-weight Qwen 3.6 series stays on Apache 2.0, with Qwen3.6-35B-A3B the open top pick. The structure mirrors a broader bet: open lower tiers seed global adoption while proprietary flagships monetise via cloud. It sets Alibaba against fully open DeepSeek V4 inside China and feeds the compute-frontier contest.

By the numbers

  • 20 May 2026, Qwen3.7-Max unveiled (proprietary).
  • 1 Jun 2026, Qwen3.7-Plus agent variant (proprietary, ~1/6 cost).
  • 1M tokens, Qwen 3.7 context.
  • Apache 2.0, licence on the open Qwen 3.6 line.

Why it matters

Alibaba is hedging the compute-frontier question: keep the frontier proprietary to drive cloud revenue, keep a strong open tier to hold developer mindshare and counter DeepSeek. The tiering shows the Chinese majors diverging on openness even as both press the agent frontier.

What to watch

  • Whether a Qwen 3.7 open-weight tier ever ships.
  • Qwen 3.7 agent benchmarks vs. DeepSeek V4 and US models.
  • Alibaba Cloud's AI revenue trajectory off the proprietary tier.

The briefing, by email