“`html
CNBC AI News, July 31st – Alibaba’s Tongyi Qianwen has unveiled its revamped reasoning model, the Qwen3-30B-A3B-Thinking-2507, marking a significant leap in both mathematical prowess and coding proficiency, according to recent reports.
The upgraded model secured an impressive score of 85.0 on the AIME25 math reasoning benchmark and achieved a score of 66.0 in the LiveCodeBench v6 coding challenge. These results position it firmly ahead of competitors, including Google’s Gemini 2.5-Flash (thinking) and Alibaba’s own previous flagship model, the Qwen3-235B.
Beyond quantitative improvements, the new model displays enhanced acumen in knowledge benchmark assessments (GPQA, MMLU-Pro). Furthermore, it demonstrates a competitive edge across a spectrum of general-purpose capabilities, including writing (WritingBench), agent functionality (BFCL-v3), multi-turn dialogue, and multi-lingual instruction understanding (MultiIF). Performance data suggests that the Qwen3-30B-A3B-Thinking-2507 provides a comprehensive advantage over established industry benchmarks.
A key feature of this release is the expanded “thinking length” afforded to the model, unlocking greater potential in processing highly complex tasks. Developers can experiment with extended deliberation times, enabling the model to better leverage its capacity for intricate logical reasoning and multi-step problem-solving.
The model is now available on open-source platforms Modeltower Community and Hugging Face. Concurrently, the Qwen Chat intelligent dialogue platform has been launched to provide users with a more streamlined and efficient conversational experience. This latest advancement signals a significant technical milestone for Tongyi Qianwen in the specialized reasoning model domain, empowering the developer community with a robust AI foundation.
“`
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/6047.html
