On July 29, the Qwen3-30B-A3B model released a new version called Qwen3-30B-A3B-Instruct-2507. This new version has achieved significant improvements in multiple key areas, marking that the model can achieve performance comparable to top closed-source models such as Gemini2.5-Flash (non-thinking) and GPT-4o, even when only activating 3B parameters in non-thinking mode.
The new version of Qwen3-30B-A3B-Instruct-2507 has made great progress in general capabilities, covering aspects such as instruction following, logical reasoning, text understanding, mathematics, science, programming, and tool usage. This improvement not only enhances the model's versatility but also makes it more intelligent and efficient when handling complex tasks. In addition, the model has made significant progress in covering long-tail knowledge in multiple languages, enabling it to better understand and generate text in various languages, which is a major advantage for global applications.
In subjective and open-ended tasks, the new model is more closely aligned with user preferences, capable of generating higher quality text and providing more helpful responses to users. This improvement makes the model more natural and human-like in interactions with users, better meeting user needs. At the same time, the ability to understand long texts has been significantly improved, reaching 256K, further enhancing the model's contextual understanding, allowing it to handle longer text inputs and better understand and generate complex text content.