The highly anticipated third generation of the open-source artificial intelligence ( AI ) model series, which promises faster processing and enhanced multilingual capabilities, was unveiled on Tuesday by Alaba Group Holding, boosting competition in an already crowded Chinese market.
According to the Qwen staff at Alibaba’s cloud computing system, the Qwen3 community consists of eight designs that range from 600 million to 235 billion, with improvements for each model. The South China Morning Post is owned by Alababe.
Parameters are used to measure the factors present during device education in AI. They serve as indicators of sophistication because larger parameter sizes normally indicate greater capacity.
In areas like training following, coding support, word technology, numerical skills, and complicated problem solving, Alibaba’s benchmark tests revealed that models like the Qwen3-235B and Qwen3-4B matched or exceeded the performance of advanced models from both domestic and international competitors – including OpenAI’s o1, Google’s Gemini, and DeepSeek’s R1.
The release of Qwen3, which was anticipated this quarter, as previously reported by the Post, is anticipated to strengthen Alibaba’s status as a leading supplier of open-source designs. Qwen is currently the largest open-source AI habitat in the world, surpassing Llama society from Facebook’s family Meta Platforms, with over 100 000 generic versions built upon it.
The Qwen group stated that” Qwen3 represents a major milestone in our development toward artificial general intelligence and unnatural superintelligence,” noting that improved pre-training and validation learning helped the new models reach a higher level of intelligence.
Qwen3 has improved abilities to comprehend and interpret directions across multiple languages, according to the staff, trained on 36 trillion tokens that span 119 languages and dialects, making the number of languages covered by Qwen2.5 trip.
Alibaba’s own AI model hosting service, ModelScope, and Microsoft’s Git Hub, the open-source AI community Hugging Face, and the Alibaba Qwen3 model family are accessible. Additionally, it has been used as the customer query definition type in the web-based Qwen robot.
People can switch between a” thinking” method, which is appropriate for complicated issues and requires longer to respond, and a “non-thinking” method, which provides quicker actions for everyday things, on all Qwen3 designs.
Alibaba’s most recent AI design was released just days after Baidu released two innovative models amid rumors about the upcoming release of DeepSeek’s R2 and R2. The development highlights the growing competition in China’s fundamental AI design market as Big Tech companies work to improve and expand their offerings.
The e-commerce giant, which is based in Hangzhou, has been increasing its investment in artificial intelligence, focusing on funding and hiring new employees to retain its edge over competitors and improve its business operations.
Alibaba pledged more than US$ 52bil  over the next three years to build Artificial system, making it the largest technology job by a personal company in China. Moreover, the organization launched a spring hiring strategy, with half of the apprenticeship positions being devoted to AI-focused positions. South China Morning Post