China’s Alibaba Group is now making its latest offering in the AI sector with its new Qwen3 model series. The Qwen3 series includes both dense and mixture-of-experts (MoE) models, with parameter counts ranging from 0.6 billion to 235 billion (in the AI field, larger parameter counts generally correlate with higher problem-solving capabilities).
“Today, we are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, general capabilities, etc., when compared to other top-tier models such as DeepSeek-R1, o1, o3-mini, Grok-3, and Gemini-2.5-Pro. Additionally, the small MoE model, Qwen3-30B-A3B, outcompetes QwQ-32B with 10 times of activated parameters, and even a tiny model like Qwen3-4B can rival the performance of Qwen2.5-72B-Instruct,” read an official blog post on the matter.
The unveiling of Qwen3 follows a series of rapid developments from Alibaba since the beginning of 2025. Just earlier this year, the company launched its Qwen2.5 series, aiming to integrate text, image, audio, and video processing within a single model. Other Chinese firms, notably Baidu with its Ernie X1 Turbo models, are similarly accelerating their AI initiatives. According to Alibaba, its latest family of models has been trained on a dataset comprising nearly 36 trillion tokens, encompassing textbooks, question-answer pairs, code snippets, and AI-generated content. This scale of training places Qwen3 firmly among the upper echelon of open-source AI developments, with some of the models available for download through platforms such as Hugging Face and GitHub.
Alibaba’s cloud division is leveraging its Singapore availability zones to deploy these services, and notes that the AI models will come with support for 119 languages and dialects. Alongside the core models, Alibaba introduced new developer-focused tools such as AI Doc for document analysis and Smart Studio for generative content creation. The company has also updated its Quark AI assistant to combine chatbot interaction with complex task execution as well. This aggressive expansion comes amid a wider trend in China’s tech sector, where companies are rapidly launching AI products at low or no cost, inspired by the success of firms like DeepSeek.
The Qwen3 models feature “hybrid” reasoning abilities, so users can choose between modes that offer quick responses to prompts, and more complex, slower “thinking” modes wherein the models offer more context and explanations to answers. “We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget,” the Qwen team stated. This approach is similar to OpenAI’s latest models and can improve both efficiency and depth of AI-generated responses. In addition to this, several Qwen3 models utilize a “mixture of experts” (MoE) architecture, which partitions tasks into smaller sub-tasks assigned to specialized mini-models. This results in improved computational efficiency as well.
In terms of performance benchmarks, Alibaba claims that its Qwen-3-235B-A22B model outperforms OpenAI’s o3-mini and Google’s Gemini 2.5 Pro on tasks like complex math and programming contests, although the largest model has not yet been made publicly available. Still, it represents a direct challenge to the dominance of American tech giants. OpenAI, Google, and Anthropic have historically benefited from greater access to high-end computing resources and regulatory support in the US. However, Alibaba’s latest family of AI models is likely to stiffly compete with OpenAI’s “o” family of models, Google’s Gemini, and other US-based AI models.
Content originally published on The Tech Media – Global technology news, latest gadget news and breaking tech news.
