Aibaba's new AI-model QWEN3 family has crossed Deepseek R1 to become the best open source model in the world. According to reports, QWEN3 did better than R1 tests that measure the abilities of AI models with open source code in areas such as language teaching, mathematics, coding and data analysis.
The Qwen3 family was launched last week by the Alibaba Cloud Database. It has eight advanced models, between 600 and 235 billion parameters. In the case of machine training, the parameters are variables until it is trained.
In accordance Livebenc The platform, an independent platform that tests major language models, had been the best open source AI-model in the world before these new tests of Deepseek R1 after it came out in January. But no longer.
Both US and Chinese companies rush to adopt Qwen 3
The rise of the Qwen3 on the livebench leaderboard shows how fast AI is developing in China. The Chinese technology industry has grown much thanks to open source tools. Alibaba's open source code method has allowed other third -party software developers to share design, improve broken links, or make the program more powerful.
However, the general Livebench results showed that Qwen3 was not as good as Openai O3, Google Gemini Pro 2.5 and Anthropici Claude 3.7, which are the best closed source AI-models in the world. Livebench says O3-Mini High, Openai's most popular AI model, was the best worldwide. Microsoft supports Openai.
For each million, it takes $ 10 to run O3. On the other hand, it is cheaper to use Qwen3 because it costs only $ 0.55 in a million ghek. As Qwen3 is cheaper and works better, many companies said they would support Alibaba's latest AI model as soon as it came out.
Huawei Technologies, Moore Threads, Cambricon Technologies and Hygon Information Technology are all chip companies that have said they support QWEN3.
Cambricon said last Tuesday that it has been successfully optimized by QWEN3 to quickly launch its graphics operating units. This was done because the developers of the Philippine AI wanted a chip made in China.
QWEN3 is also used in hyperbolic and fireworks cloud data processing services. The American smart producers Nvidia and Intel have begun to support Qwen3.
Many major dual dual centers in China, such as Beijing, Shanghai, Hangzhou and Hubei, Jilin and Northwest-Shanxi provinces, have also said that they use Alibaba's third-generation Qwen AI models. The supercomputer network in China has also accepted Qwen3. This network connects over 20 data center in 20 cities in 14 provinces.
Anthropic CEO says that Deepseek was “a little over blown”
At the company's event, the co -founder of Antropic, a company that compiled Claude AI, said that Deepseek is still “six to eight months behind, where US border companies are located”. He also said that the recent Buzz around the Chinese start-up was “maybe a bit blown”.
Deep Got attention to the world in late December 2024 and early January 2025, sharing two AI model, V3 and R1, with open source code. These models were made of a small part of the cost and the calculation power that large technology companies usually need for LLM projects.
It is unclear when Deepseek releases the next generation of its models. The company in Hangzhou quietly released its 671 billion parameter grandfather-v2 late AprilTo. It was an update for this specialized model to handle math evidence. However, he has not said anything about the progress of his long -awaited R2 reasoning model.
Cryptopolitan Academy: Soon – a new way to earn a passive income with a defi in 2025. For more information
