China's DeepSeek Ups The Heat In OpenAI Rivalry, Upgrades V3 Model Improving Coding And Reasoning Capability

On Tuesday, Chinese AI startup DeepSeek released its upgraded DeepSeek-V3-0324 model, enhancing reasoning and coding abilities.

Available on Hugging Face, the model shows improved performance in technical benchmarks, intensifying competition with U.S. leaders like OpenAI and Anthropic.

The updated model has shown improvements across multiple benchmarks, particularly scoring 59.4 on the American Invitational Mathematics Examination, up from 39.6 in its predecessor, according to a report by the South China Morning Post.

Also Read: Alibaba’s Tsai Sounds Alarm On Signs Of AI Bubble In US: ‘People Are Building Data Centers On Speculation’

It also gained 10 points on LiveCodeBench, reaching a score of 49.2. The new model, with 685 billion parameters, uses an MIT software license, unlike DeepSeek V3, which has 671 billion parameters and a commercial license.

Earlier this year, DeepSeek’s R1 model disrupted American tech supremacy, sparking debates about Big Tech’s significant investments in large language models and data centers.

R1 made waves with its performance and lower costs, but analysts reportedly believe DeepSeek’s biggest impact is encouraging the use of open-source AI models, CNBC reports. This shift has been a key factor in the company’s influence on the industry.

Wei Sun, principal analyst at Counterpoint Research, told CNBC that DeepSeek’s success shows open-source strategies drive faster innovation and wider adoption, with many companies adopting the model. She also mentioned that R1 is influencing China’s AI scene, prompting major firms like Baidu to open-source their own LLMs in response.

Recently, Kai-Fu Lee, the former head of Alphabet Inc.’s GOOG GOOGL Google China and founder of AI startup 01.AI, said that the rise of open-source AI models like DeepSeek has exposed an existential risk to OpenAI’s business—and he’s pivoting his company accordingly.

He questioned the long-term sustainability of OpenAI’s business model, especially when competing against open-source projects that offer similar quality at a fraction of the cost.

Tim Wang, managing partner at Monolith Management, shared in an interview with CNBC that models from companies like DeepSeek have been powerful enablers in China, showing how progress can be made with fewer resources.

He noted that open-source models have reduced costs, allowing for product innovation—an area where Chinese companies excel.

Wang compared this development to the “Android moment,” when Google’s decision to make its operating system’s source code available sparked innovation in the app ecosystem.

He added that the perception of China being 12 to 24 months behind the U.S. in AI has now shifted to just 3 to 6 months.

This shift in China’s AI landscape, driven by open-source models like DeepSeek, is changing global competition and challenging traditional business models.

Read Next: