DeepSeek has released a new version of its R1 AI reasoning model without making a formal announcement. The model was quietly uploaded to Hugging Face where it drew immediate attention from the AI community. DeepSeek shocked the tech world earlier this year with its fast development speed and low-cost structure. 

“DeepSeek’s latest upgrade is sharper on reasoning, stronger on math and code, and closing in on top-tier models like Gemini and O3.”

Its free open-source R1 model previously outperformed big names like Meta and OpenAI. The upgraded version builds on that success and ranks just behind OpenAI's o4-mini and o3 on LiveCodeBench. This benchmark site compares AI models across different skill sets. Experts say DeepSeek’s new model has improved reasoning skills and better math and coding capabilities.

China’s AI Race Gains Speed despite Export Limits

The upgrade has also reduced hallucinations which are incorrect or made-up responses. Adina Yakefu from Hugging Face said the new model proves DeepSeek is not only catching up but competing. DeepSeek’s rise has put pressure on U.S. tech companies and even led to stock value dips in firms like Nvidia. 

While those values have recovered DeepSeek's progress remains impressive. China’s AI industry is advancing despite U.S. restrictions on chip exports. Major Chinese companies like Baidu and Tencent are also finding ways to make their models more efficient. Nvidia CEO Jensen Huang said it was wrong to assume China cannot develop its own AI chips. 

He pointed out that China already has a powerful AI sector and a large pool of researchers. DeepSeek is now seen as proof that the global AI race is no longer led only by the U.S. The quiet release of its upgraded model signals growing confidence in its technology and strategy.