On the evening of February 3rd, Alibaba took the tech world by storm by open-sourcing its latest-generation agent programming model, Qwen3-Coder-Next. Boasting a total of 80 billion parameters, this groundbreaking model cleverly activates only 3 billion parameters during each inference process. This strategic design substantially cuts down on inference costs, making it a cost-effective solution in the realm of programming models.

Remarkably, its agent programming capabilities stand toe-to-toe with those of top-tier open-source models like DeepSeek-V3.2 and GLM-4.7. What sets Qwen3-Coder-Next apart is its innovative approach to agent training expansion. Through this cutting-edge method, the model gains the remarkable ability to 'think' while programming. This breakthrough opens up exciting new avenues for smaller models, enabling them to tackle long-context and complex agent tasks with unprecedented ease and efficiency.