CEO's Column
Search
More
Foundation Models

Alibaba Unveils Qwen3 AI Models

ByRishabh Srihari
2025-04-29.13 days ago
Alibaba Unveils Qwen3 AI Models
Alibaba Unveils Qwen3 AI Models, Image Credit: Freepik AI

Chinese tech giant Alibaba has introduced Qwen3, a family of advanced AI models that it claims can rival, and sometimes outperform, top-tier models developed by Google and OpenAI.

Accessible and Powerful AI Models

The Qwen3 models, which range in size from 0.6 billion to 235 billion parameters, are designed to address a wide array of tasks, from simple queries to more complex problem-solving. Alibaba has made most of these models available under an open license on popular AI development platforms like Hugging Face and GitHub, providing access for developers worldwide.

Parameters, the building blocks of AI models, are critical in determining a model's problem-solving capabilities. The larger the number of parameters, the more effective a model generally is. With its vast range, Qwen3 brings high-end AI performance to a broader audience.

Hybrid Design for Efficiency and Flexibility

Alibaba describes the Qwen3 models as "hybrid" systems, combining fast, responsive capabilities with the ability to reason through complex problems. This reasoning ability allows the models to fact-check their own answers, akin to OpenAI’s o3, though with higher latency. The Qwen team has also introduced a feature that allows users to control the "thinking budget," enabling task-specific configuration to optimize performance.

Some of the models employ a Mixture of Experts (MoE) architecture, which breaks tasks into subtasks and delegates them to specialized expert models. This results in greater computational efficiency, particularly when answering more specific queries.

Also read: Alibaba Set to Launch Qwen 3 AI Model Amid Intensifying Competition with DeepSeek

Competing with Industry Giants

Despite some limitations, Qwen3 has demonstrated impressive performance. The largest public model, Qwen3-32B, outperforms OpenAI’s o1 model in benchmarks like LiveCodeBench, a coding test. The largest model, Qwen-3-235B-A22B, has even surpassed Google’s Gemini 2.5 Pro and OpenAI’s o3-mini in certain tests, including math challenges and problem-solving assessments on platforms like Codeforces. However, Qwen-3-235B-A22B is not yet publicly available.

Qwen3 is also noted for its excellent tool-calling abilities and its capacity to follow instructions and handle various data formats. Available for download and through cloud providers like Fireworks AI and Hyperbolic, it is being adopted quickly by developers.

Implications for the AI Landscape

Industry experts, such as Tuhin Srivastava, co-founder of Baseten, highlight that the rise of open models like Qwen3 is narrowing the gap with closed-source systems from companies like OpenAI. Despite the US continuing to restrict chip sales to China, state-of-the-art open models like Qwen3 will still play a significant role in the global AI landscape, as businesses continue to balance the development of in-house tools and external solutions.

Alibaba’s release of Qwen3 marks a significant milestone in the AI arms race, with the potential to reshape the competitive dynamics between Chinese and Western companies in the AI space.

Related Topics

LLMsLarge Language Models (LLMs)

Subscribe to NG.ai News for real-time AI insights, personalized updates, and expert analysis—delivered straight to your inbox.