There’s a new candidate making waves in the AI field: QwQ-32B-Preview.
This so-called “inference” AI model has been compared to OpenAI o1 and is one of the few models available for download under a permissive license. This is a huge benefit for developers and researchers who are keen to experiment.
Built by Alibaba’s Qwen team, the QwQ-32B-Preview is anything but lightweight. Packed with 32.5 billion parameters (think of them as building blocks for problem-solving ability), it can handle prompts of up to 32,000 words. It’s longer than other novels! In testing, it has shown better performance than OpenAI o1-preview and o1-mini on benchmarks such as AIME and MATH. For context, AIME uses other AI models to evaluate performance, while MATH is a collection of word problems.
But this model is not just about math problems and logic puzzles. What makes them unique is the way they approach tasks. QwQ-32B-Preview helps you plan ahead, fact-check your work, and avoid common AI mistakes. Of course, it’s not perfect. Alibaba has acknowledged issues such as language switching, occasional loops, and difficulty with “common sense” reasoning. Still, this represents a step toward more intelligent AI systems.
You can access QwQ-32B-Preview. It can be run or downloaded via Hugging Face. However, like other Chinese-developed AIs, it operates within regulations. That means carefully avoiding politically sensitive topics in order to follow China’s rules and ensuring that China is consistent with “socialist core values.”
Alibaba is not the only company active in this field. Meta’s Llama 3.1 is another open source option, but it takes a different approach by focusing on generative AI rather than inference. Both models are innovative, but the QwQ-32B-Preview falls into the reasoning category, specializing in problem solving with what the company describes as a human-like approach.
Competition in AI within China is intensifying. Companies such as DeepSeek, Shanghai AI Lab, and Kunlun Tech have entered the inference AI race and are rapidly releasing their models. For example, DeepSeek’s r1 claims to perform better than OpenAI’s o1 on half of the benchmark tests, especially math and programming. Shanghai AI Lab’s InternThinker takes a structured approach to problem solving that incorporates steps such as understanding a query, recalling knowledge, planning a solution, and reflecting on the answer.
The surge in activity highlights how quickly Chinese companies are catching up with American tech giants. Xu Liang, an AI entrepreneur from Hangzhou, summed it up as follows: Research is making progress for Chinese technology companies. ” The releases of the QwQ-32B-Preview and its competitors show how much ground they cover.
But this is about more than just catching up. Reasoning AI changes the way models are designed and used. Unlike older AI systems that generate answers using brute force, reasoning models like QwQ-32B-Preview aim to mimic human problem solving. This approach not only improves efficiency in complex tasks, but also expands potential use cases, such as tackling advanced mathematics and providing detailed financial advice.
Whether it’s solving puzzles, reasoning through complex problems, or expanding what open source AI can accomplish, one thing is clear: This means that the evolution of AI is accelerating. Please buckle up. This is just the beginning.
(Photo courtesy of Unsplash)
See also: Alibaba Cloud overhauls AI partner initiatives
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event will be co-located with major events such as Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Learn about other upcoming enterprise technology events and webinars from TechForge here.
