Alibaba has introduced Qwen3, the latest in its open-sourced large language model series, marking a new benchmark for AI innovation. The Qwen3 series features six dense models and two Mixture-of-Experts (MoE) models, offering developers flexibility to build next-generation applications across various platforms including mobile devices, smart glasses, autonomous vehicles, and robotics.

All Qwen3 models are now open-sourced and globally available. These include dense models (0.6B, 1.7B, 4B, 8B, 14B, and 32B parameters) and MoE models (30B with 3B active, and 235B with 22B active).

Qwen3 marks Alibaba's debut of hybrid reasoning models, combining traditional LLM capabilities with advanced, dynamic reasoning. These models can switch between thinking mode for complex tasks like mathematics, coding, and logical deduction, and non-thinking mode for fast, general-purpose responses. Developers accessing Qwen3 through API gain granular control over thinking duration, optimizing the balance between intelligent performance and compute efficiency. The Qwen3-235B-A22B MoE model significantly lowers deployment costs compared to other state-of-the-art models.

Trained on a dataset of 36 trillion tokens, double that of its predecessor Qwen2.5, Qwen3 delivers advancements in reasoning, instruction following, tool use, and multilingual tasks. It supports 119 languages and dialects, with leading performance in translation and multilingual instruction-following. Qwen3 natively supports the Model Context Protocol (MCP) and robust function-calling, leading open-source models in complex agent-based tasks. It surpasses previous Qwen models in mathematics, coding, and logical reasoning benchmarks and delivers more natural creative writing, role-playing, and multi-turn dialogue experiences.

Qwen3 models achieve top-tier results across industry benchmarks like AIME25, LiveCodeBench, BFCL, and Arena-Hard. The hybrid reasoning model was developed using a four-stage training process: long chain-of-thought (CoT) cold start, reasoning-based reinforcement learning (RL), thinking mode fusion, and general RL.

The launch of Qwen3 comes amid rising competition in China’s AI sector, spurred by the success of local startup DeepSeek and the release of Baidu's Ernie 4.5 Turbo and Ernie X1 Turbo models. Alibaba's newest release merges conventional AI functions with advanced dynamic reasoning, creating a platform for app and software developers. Alibaba had previously released its Qwen 2.5-Max model in late January, claiming superior performance.

Qwen3 models are available for download on Hugging Face, GitHub, and ModelScope, and can be explored on chat.qwen.ai. API access will soon be available through Alibaba's AI model development platform Model Studio. Qwen3 also powers Alibaba's flagship AI super assistant application, Quark. Since its debut, the Qwen model family has attracted over 300 million downloads worldwide, with developers creating more than 100,000 Qwen-based derivative models on Hugging Face.