Alibaba has launched Qwen 3, a new generation of artificial intelligence models that possess advanced “hybrid” capabilities. These models are great at tackling complicated problem-solving tasks. Besides long-form answers, they can answer shorter, more straightforward questions quickly, so they’re flexible tools with many possible uses. The press release noted Alibaba’s plans to continue advancing AI capabilities and increasing its availability in over 100 languages.
The Qwen 3 models were carefully trained on an enormous set of data. This dataset includes textbooks, question-answer pairs, computer code snippets and more, totaling almost 36 trillion tokens. It is this transformative training that empowers the models to serve as many as 119 languages, making them usable and effective in a variety of environments. The Qwen 3 supports extensive parameters from 0.6 billion to 235 billion. This flexibility to prioritize different user needs and computational capabilities is highly valuable.
Of these released models, the largest public variant, Qwen3-32B, is notable for its strong performance comparable to that of proprietary AI systems. In fact, it has already outperformed OpenAI’s o1 model on a variety of performance benchmarks, such as LiveBench, demonstrating its superior capability to produce high-quality responses. The Qwen-3-235B-A22B model exhibited outstanding capabilities in mathematical reasoning. It has surpassed OpenAI’s o3-mini on the AIME math benchmark and BFCL reasoning exam.
Moreover, Qwen-3-235B-A22B has demonstrated outstanding performance on Codeforces, outperforming both OpenAI’s o3-mini and Google’s Gemini 2.5 Pro. This point of accomplishment underscores Alibaba’s progress in AI reasoning and problem-solving capabilities. Filling that very role, the Qwen team previously underscored the model’s impressive capacity when they wrote,
“We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget.”
Alibaba says that Qwen 3 has strong reasoning skills. It’s exceptional for its tool-calling capabilities and its remarkably literal follow-through with user commands. At the moment, the models are only available through Fireworks AI and Hyperbolic, cloud providers. That said, the most advanced variant, Qwen-3-235B-A22B, is currently not publicly available.
The release of Qwen 3 comes at an interesting time, as the global AI race continues to play out in increasingly rapid succession. According to industry experts, businesses are more interested than ever in both bespoke solutions and off-the-shelf models from established companies. Tuhin Srivastava remarked on this trend, saying,
“It reflects the reality that businesses are both building their own tools [as well as] buying off the shelf via closed-model companies like Anthropic and OpenAI.”
As geopolitical factors influence the AI market, particularly regarding U.S.-China trade relations, Srivastava pointed out the significance of open models like Qwen 3:
“The U.S. is doubling down on restricting sales of chips to China and purchases from China, but models like Qwen 3 that are state-of-the-art and open […] will undoubtedly be used domestically.”