Alibaba Releases Qwen3, New AI Models with Hybrid Reasoning Capabilities

On April 28, 2025, Chinese tech giant Alibaba introduced Qwen3, a new family of AI models that aim to compete with top models from Google and OpenAI. The Qwen3 models come in a variety of sizes, ranging from 0.6 billion to 235 billion parameters. Parameters are key indicators of a model’s problem-solving capabilities, with larger models generally showing better performance. Alibaba describes Qwen3 models as ‘hybrid’ AI, which means they can both reason through complex problems and handle simple tasks quickly. This dual approach allows the models to fact-check themselves, similar to OpenAI’s models, though with slightly higher latency. Qwen3 also incorporates a mixture of experts (MoE) architecture, which increases computational efficiency by breaking tasks into smaller parts. With support for 119 languages, Qwen3 was trained on a massive dataset of nearly 36 trillion tokens, further enhancing its capabilities. The Qwen3 models excel in tool-calling, following instructions, and copying specific data formats. Alibaba claims that these improvements make Qwen3 a strong competitor, though it does not surpass the latest models from OpenAI and Google in every metric. Despite the competition, Qwen3 represents a significant leap forward for AI models from China, and its open release has the potential to influence the global AI landscape. The models are available for download from platforms like Hugging Face and GitHub, and can also be accessed through cloud providers such as Fireworks AI and Hyperbolic.