QWEN3 is Alibaba’s debut into the so-called “hybrid inference model,” combining traditional LLM features with “advanced and dynamic inference.”
SOPA Images | Lightrocket | Getty Images
Alibaba On Tuesday, it released the next generation of large-scale open source language models.
in Blog postThe Chinese tech giant said QWEN3 promises to improve inference, teaching, tools use, and multilingual tasks, comparable to other top tier models such as DeepSeek’s R1 in several industry benchmarks.
The LLM series includes eight variations across a variety of architectures and sizes, providing developer flexibility when building AI applications for edge devices such as mobile phones using QWEN.
QWEN3 is also the debut of Alibaba’s so-called “hybrid inference model,” combining traditional LLM features with “advanced and dynamic inference.”
According to Alibaba, such models can seamlessly transition between “thinking modes” for complex tasks such as coding and “non-thinking modes” for faster, general-purpose responses.
“In particular, the QWEN3-235B-A22B MOE model significantly reduces deployment costs compared to other cutting-edge models and strengthens Alibaba’s commitment to accessible, high-performance AI,” said Alibaba.
The new model is already freely available to individual users on platforms such as Face and Github Hugs, as well as Alibaba Cloud’s web interface. QWEN3 is also used in the power of Quark, Alibaba’s AI assistant.
Advances in China’s AI
AI analyst told CNBC that QWEN3 is a serious challenge for the counterparts of China’s Alibaba, as well as for the US industry leaders.
In a statement to CNBC, Wei Sun, principal analyst for artificial intelligence at Counterpoint Research, said the QWEN3 series is a “significant breakthrough” for several features that refer to “not only best-in-class performance, but also the ability to apply models.”
These features include QWEN3’s hybrid thinking mode, multilingual support covering 119 languages and dialects, and open source availability.
Open source software generally refers to software in which source code is freely available on the web, which may be altered and redistributed. Earlier this year, Deepseek’s open source R1 model rocked the world of AI, and quickly got it right away. Catalysts for adoption of Chinese AI space and open source models.
“The release of Alibaba’s Qwen 3 Series further highlights the strong ability of Chinese labs to develop highly competitive, innovative and open source models despite pressure from strengthening US export controls.”
According to Alibaba, Qwen has already become one of the most widely adopted in the world Open Source AI Model SeriesIt has attracted over 300 million downloads around the world, and has embraced over 100,000 derivative models.
Wang said the adoption could continue with QWEN3, adding that its performance claims could still be the best open source model globally, although it still lies behind the world’s most cutting edge models, such as Openai’s O3 and O4-Mini.
Chinese competitors like Baidu are rushing to release new AI models, including planning to shift to a more open source business model after the advent of Deepseek.
Meanwhile, Reuters It has been reported In February, Deepseek accelerated the launch of its successor to R1, citing anonymous sources.
“In the broader context of American and Chinese AI races, the gap between the US and Chinese labs is narrowing. They may argue for even just weeks, up to months,” Wang said.
“With the latest release of Qwen 3 and the upcoming release of Deepseek’s R2, this gap is unlikely to grow and may even continue to shrink.”
