Multimodal AI optimized for real-world coding agents
Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3
Qwen3 provides dense and MoE model variants (0.6B to 235B parameters) with both thinking and non-thinking modes, enabling developers to select the right model size for reasoning tasks like mathematics, coding, and logical inference while maintaining full control over deployment and customization
Qwen3 demonstrates leading performance in agent capabilities with precise external tool integration in both thinking and non-thinking modes, enabling developers to build autonomous systems that can plan and execute complex multi-step tasks
Qwen3 supports 100+ languages and dialects with strong multilingual instruction following and translation capabilities, available as open-source weights that can be deployed on-premises or in private cloud environments
Qwen3 offers multiple sizes (0.6B, 1.7B, 4B, 8B, 14B, 32B, 30B-A3B, 235B-A22B) and both dense and Mixture-of-Experts architectures, allowing users to balance reasoning quality, latency, and computational cost for their specific use case
Qwen3-2507 supports 256K-token long-context understanding natively and is extendable up to 1 million tokens, enabling applications that require processing of lengthy documents, multi-turn conversations, and large codebases