Six months ago, Chinese AI labs were in a race to publish open-weight models. Now they're all running the same playbook: hold the release, promise it's coming soon, say the model needs more work first.
Minimax-m2.7, GLM-5.1, GLM-5-turbo, GLM-5v-turbo, Qwen3.6, and Mimo-v2-pro are all sitting unreleased as of early April 2026. Each lab has issued similar statements: the model exists, it's being refined, it will be open-sourced eventually. The language is strikingly uniform across organizations that normally compete aggressively with each other.
Three Possible Explanations
Regulatory pressure. China's AI governance rules tightened in 2025, and there's an open question about what Chinese labs can release internationally without review. A coordinated pause would make sense if all labs received the same guidance from regulators - a message doesn't have to be a formal regulation to be effective.
Strategic recalibration. DeepSeek's open-source releases generated enormous global attention but also handed Western labs free access to China's best research. Minimax, Zhipu (the company behind GLM), Alibaba (Qwen), and Mimo may have all reached the same conclusion: open-sourcing flagship models no longer makes competitive sense. If one lab shifted on this, others would follow fast.
Actual coordination. Less likely, but not impossible. Chinese tech companies operate in an environment where government bodies can convene them informally. A collective instruction to pause international open releases wouldn't necessarily appear in any public document.
What makes this pattern notable is the timing. These aren't labs on staggered schedules who happened to hit delays in the same quarter. These are separate teams, at separate companies, building separate architectures - all pausing in roughly the same 60-day window and using nearly identical public messaging.
What This Means for People Who Build With Open Models
The global open-source AI community has leaned heavily on Chinese lab releases over the past 18 months. Qwen models became standard baselines for fine-tuning (training a model on new data to specialize it for a specific task) and running AI locally without a cloud subscription. If this pause signals a permanent strategy shift, the supply of freely available frontier-level models will shrink.
Western open-source alternatives exist - Meta's Llama series is the most prominent - but Chinese labs have been releasing models at price-to-performance ratios Meta hasn't consistently matched. A sustained withdrawal would be felt by developers who rely on these models for cheap, capable local inference (running AI on your own hardware instead of paying per API call).
The charitable read is that all four models are genuinely being improved and will ship within weeks. That's plausible. But the same story, from four separate organizations, at the same moment, with near-identical wording, is a pattern that deserves more than a shrug.