China’s DeepSeek Quietly Releases Upgraded AI Reasoning Model, Narrowing Gap with Global Leaders

Chinese AI startup DeepSeek, which rattled global markets earlier this year with its powerful open-source reasoning model, has quietly released an upgraded version of its flagship system, DeepSeek R1, without a formal announcement. The update was made public on Hugging Face, where the model has rapidly gained attention.

The original DeepSeek R1 stunned the industry by outperforming models from Meta and OpenAI, despite being developed in a fraction of the time and cost. Its success triggered a market shakeup, raising concerns about overinvestment by U.S. AI giants and briefly erasing billions in market value from firms like NVIDIA.

Now, the latest version of DeepSeek R1 is showing significant progress in logical reasoning, math, and coding performance — ranking just behind OpenAI’s o4-mini and o3 on LiveCodeBench, a key benchmark for reasoning models.

Adina Yakefu, an AI researcher at Hugging Face, said this version shows major improvements in inference and hallucination reduction. While adding that DeepSeek isn’t just catching up — it’s competing directly with top-tier models like Gemini and O3.

The quiet release underscores China’s continued momentum in AI, despite U.S. export restrictions on chips and advanced tech.