On July 30th, the Alibaba Tongyi Qianwen team announced the latest iteration of the Qwen3 series, the Qwen3-30B-A3B-Instruct-2507 model. This release marks a significant advancement in the realm of non-thinking models, as it boasts the ability to match the performance of closed-source models such as Gemini 2.5-Flash (non-thinking) and GPT-4o across multiple core capabilities with the activation of just 3B parameters. Furthermore, the Qwen3-30B-A3B-Instruct-2507 model has achieved groundbreaking progress in multilingual support, user preference alignment, and long text processing.