Qwen 3.7 Max and the China frontier leadership — what the GPQA Diamond crown means for the open-vs-closed competitive frame

Alibaba's Qwen 3.7 Max scored 92.4 on GPQA Diamond — beating Claude Opus 4.6. The capability-gap rhetoric that dominated 2024-2025 ("open weights trail proprietary by 6-12 months") no longer matches the data. Nine frontier-class open-weight models shipped in six weeks. The frame has shifted from chase to race.

The cadence is the substance. qwen 3 7 max may 20 gpqa diamond beats claude opus documents the score. zai glm 5 1 may 2026 open weight release is just one of the nine open-weight releases in the April-May 2026 window: Moonshot Kimi K2.6, Z.ai GLM-5.1, DeepSeek V4 Pro and V4 Flash, Xiaomi MiMo-V2.5-Pro, MiniMax M2.7, Google Gemma 4, Alibaba Qwen 3.6 then 3.7 Max, Ant Group inclusionAI Ring-2.6-1T. Six weeks, nine frontier-class drops.

The licensing matters as much as the benchmarks

Qwen 3.7 Max ships under Apache-2.0. Kimi K2.6 ships under MIT. Gemma 4 ships under Apache-2.0. Enterprise deployment teams evaluating frontier reasoning today can hit Opus-class numbers on infrastructure they fully control with permissive commercial licensing. The compliance and audit-trail tradeoff that historically pushed regulated buyers toward proprietary APIs has flipped — running the model on-prem with verified weights is now a feature, not a downgrade.

What this does to closed-lab pricing

If GPQA Diamond 92+ is available open-weight, the pricing power on Claude Opus tier contracts depends on capabilities OPEN models can't match — agent integration, deployment tooling, audit trails, multi-modal reasoning, ecosystem lock-in. anthropic project glasswing tiered frontier access is partly a response to this — productizing capabilities that aren't just "the smartest model." Closed labs are being forced to compete on the surrounding system, not just the weights.

The China-vs-rest-of-world frame

Seven of the nine April-May open-weight releases came from Chinese labs (Moonshot, Z.ai, DeepSeek, Xiaomi, MiniMax, Alibaba, Ant Group). The geopolitical implications matter: export controls intended to slow Chinese AI capability are not visible in benchmark trajectories. The cohort is shipping faster than the US closed labs and competitive on capability. That's a frame shift policymakers haven't fully internalized yet.

Understanding AI — The best Chinese open-weight models vs the strongest US rivals → · Qwen — Qwen 3.7 Max release notes →