Alibaba Introduces Qwen3 AI Models to Compete with OpenAI and Google
When OpenAI and Gemini are leading the AI model industry, we have another AI model to compete with them. Alibaba has officially launched its new family of AI models called Qwen3. According to the company, these models not only match but in some cases outperform leading models.
The Qwen3 lineup is quite broad, offering models ranging from 0.6 billion parameters up to a massive 235 billion parameters. The number of parameters shows how well a model can understand and solve problems. The higher numbers show that the model is more capable and advanced.
Alibaba Introduces Qwen3 AI Models to Compete with OpenAI and Google
Most of the Qwen3 models are available for download under an open license. You can find them on platforms like Hugging Face and GitHub. This open approach places Alibaba among a growing number of companies that are making powerful AI tools more accessible.
According to Alibaba, the Qwen3 models are hybrid models. They can either quickly answer simple queries or take more time to “reason” through complex problems. This flexibility allows users to balance between speed and deeper problem-solving, depending on their needs.
“We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget,” Alibaba’s Qwen team explained in a blog post.
Some of the Qwen3 models also use a Mixture of Experts (MoE) architecture. MoE models break tasks into smaller parts and assign them to specialised sub-models. This approach can reduce computing power requirements while maintaining high performance.
In terms of capabilities, Qwen3 supports 119 languages. The company used textbooks, code snippets, AI-generated material, and large sets of questions and answers to train the models, containing nearly 36 trillion tokens.
The release of Qwen3 also increases the competition between Chinese and American AI models. U.S.-based companies like OpenAI are now under pressure to stay ahead. In the meantime, the US government is also tightening regulations to limit China’s access to the latest AI chips.
Performance-wise, Qwen3 models have shown impressive results. The largest model, Qwen3-235B-A22B, outperformed OpenAI’s 3o-mini and Google’s Gemini 2.5 Pro in coding and math benchmarks, including tests like Codeforces and AIME. However, this flagship model is not yet publicly available.
The biggest public model, Qwen3-32B, is still very competitive. It has outperformed models like OpenAI’s 1o on coding tests such as LiveCodeBench.
See Also: OpenAI Launches Lighter Version of Deep Research for Free Users
Alibaba also highlights Qwen3’s strong abilities in tool-calling, instruction following, and data format replication. Besides downloading the models, users can also access Qwen3 through cloud platforms such as Fireworks AI and Hyperbolic.
Industry experts see Qwen3 as a major step forward for open AI models. Tuhin Srivastava, CEO of AI cloud hosting company Baseten, commented that open models like Qwen3 are now keeping pace with closed systems from companies like OpenAI and Anthropic.
Despite increasing restrictions on chip sales, Alibaba’s Qwen3 demonstrates that innovation in AI is moving fast, and not limited to just a few major players.
