Alibaba introduces and open-sources Qwen3, China's first hybrid reasoning AI model

Alibaba introduces Qwen3, China's pioneering hybrid AI model.

: Alibaba has unveiled Qwen3, marking China's entrance into hybrid reasoning AI with a model that introduces both fast and slow thinking modes, facilitating cost-efficient computations. The Qwen3 series ranges from fine-tuned variations like Qwen3-30B-A3B to pre-trained models, offering flexible task responses. Two notable open-source models are Qwen3-235B-A22B, boasting over 235 billion parameters, and a more lightweight variant with 30 billion parameters. With capabilities in coding, math, and general reasoning, Qwen3 is designed to rival global names like DeepSeek-R1 and Grok-3.

Alibaba's recent launch of Qwen3 marks a significant milestone in AI development as it becomes China's first hybrid reasoning AI model. The model integrates fast and slow thinking modes, a strategic design aimed at enhancing operational efficiency by reducing computational costs. This innovative AI model, introduced on April 29, exemplifies Alibaba's commitment to pushing technical boundaries in artificial intelligence. Models like Qwen3-30B-A3B that are fine-tuned for specific tasks, and its pre-trained base, present users with versatile options across multiple platforms.

Qwen3 distinguishes itself with its two reasoning modes, allowing users the flexibility to toggle between detailed, step-by-step solutions or quicker, more streamlined responses. This adaptability is particularly beneficial when considering varying complexities of tasks, showcasing Qwen3's balance between speed and intelligence. Such a flexible framework aids in optimizing performance tailored to individual user requirements.

Alibaba Cloud has also made significant strides by open-sourcing the Mixture-of-Experts (MoE) models, including their flagship Qwen3-235B-A22B, which comprises over 235 billion parameters. This model not only illustrates Alibaba's capabilities but directly competes with globally recognized models such as DeepSeek-R1, 01.AI’s o1 and o3-mini, Grok-3, and Gemini 2.5 Pro. By making such advanced AI technology publicly accessible, Alibaba seeks to foster greater innovation and collaboration within the AI research community.

In addition to the flagship model, a lighter version, Qwen3-30B-A3B, offers a combination of 30 billion total parameters with a subset of 3 billion active parameters. This lightweight model caters to more resource-constrained environments while still achieving high performance in specific use cases.

Through these groundbreaking releases, Alibaba not only enhances its stature in the AI domain but also broadens the horizon for future advancements. By encouraging open collaboration, the company underscores the importance of shared learning and improvement in technology. Alibaba's initiative is a pivotal step in the international AI arena, setting a precedent for technological growth through innovation and accessibility.

Sources: Alibaba Cloud, TechNode

AICloudinnovation