Just now, Alibaba released and open-sourced Qwen3, and its benchmark indicators are quite impressive!
Alibaba's latest masterpiece, the Qwen3 series of models, is released with amazing performance, leading the new trend of open source!
Core content:
1. The performance of the Qwen3 series of models is comparable to the international top AI models, showing excellent competitiveness
2. The flagship and small models have excellent performance, support 256K context length, and multi-modal expansion
3. Some model weights are open source, providing enterprise-level security customization support, and QwenChat function online experience
In the early morning of April 29, Alibaba officially released the Qwen3 series of models. Compared with top models such as DeepSeek-R1, OpenAI's o1 and o3-mini, Musk's Grok-3, and Google's Gemini-2.5-Pro, it showed extremely competitive results.
The benchmark indicators of Qwen3 are quite good. Among them, the flagship model Qwen3-235B-A22B performs well, while the small MoE model Qwen3-30B-A3B has only 10% of the number of activated parameters of Qwen3-32B, but has better performance.
In addition, the Qwen3-4B small model can also match the performance of the Qwen2.5-72B-Instruct, demonstrating its powerful lightweight application capabilities , which is very important to the open source community! It can be said that Alibaba has taken over the torch of mainstream open source model producers from Meta.
The Qwen3 series supports context lengths up to 256K, covers both reasoning and non-reasoning tasks, and continues the multimodal expansion capabilities, supporting text, image, audio and video processing, suitable for complex tasks and a variety of application scenarios. Some model weights will be open source, while providing enterprise-level security and customized deployment support.
