Qwen3 Open Source: Chinese Models Enter the Hybrid Reasoning and Agent Stage
The Qwen team released the Qwen3 series of open models, supporting thinking/non-thinking switching and multilingual capabilities.
Domestic open models are rapidly absorbing the reasoning-time compute and MoE approaches, forming a complete matrix of cost, size, and deployment options.
Hybrid reasoning allows the same model to switch thinking budgets per task, while MoE increases parameter capacity and inference efficiency.
The technology gap between Chinese models and global frontiers is narrowing, with competition shifting to ecosystems, products, computing power, and overseas distribution.
Model supply in the domestic To D market becomes more abundant, with value continuing to shift to toolchains and industry data.
Observing agent tool usage, overseas adoption, enterprise deployment, and next-generation flagship models.