Alibaba has released Qwen3, a new family of large language models that compete with leading AI systems from OpenAI and Google. The lineup includes two mixture-of-experts (MoE) models and six dense models, with parameters ranging from 0.6 billion to 235 billion. According to benchmarks shared by Alibaba, the flagship Qwen3-235B-A22B model outperforms DeepSeek R1 and OpenAI’s o1 on several key metrics, while approaching the capabilities of Google’s Gemini 2.5-Pro.
A distinctive feature of Qwen3 is its “hybrid thinking” approach, allowing users to toggle between two modes: a “Thinking Mode” for step-by-step reasoning on complex problems, and a faster “Non-Thinking Mode” for simpler queries. This functionality can be controlled with prompts like “/think” and “/no_think” or through interface buttons.
The models offer extensive language support, covering 119 languages and dialects across major language families. Alibaba trained Qwen3 on approximately 36 trillion tokens, double the dataset size of its predecessor Qwen2.5, using web content, PDF documents, and AI-generated synthetic data.
Technical capabilities and availability
Most of the Qwen3 models are available under the Apache 2.0 license, making them suitable for commercial use without usage-based restrictions. The models can be deployed through frameworks like SGLang and vLLM, which provide OpenAI-compatible endpoints, or used locally with tools such as Ollama, LMStudio, and llama.cpp.
The open-source nature of these models allows engineering teams to:
- Configure OpenAI-compatible endpoints to use Qwen3 models within hours
- Run private fine-tuning without sending proprietary data to third parties
- Prototype on laptops using smaller models and scale to larger implementations without rewriting prompts
- Keep all prompts and outputs on-premises for security and compliance
Industry observers note that Qwen3 represents another step in the trend of open-source models keeping pace with proprietary systems from companies like OpenAI and Anthropic. The release occurs amid increasing U.S. restrictions on chip sales to China, highlighting the ongoing global competition in AI development.
Sources: Qwen Team, VentureBeat, TechCrunch