Beijing, August 28, 2025 — Chinese artificial intelligence startup DeepSeek has unveiled its upgraded large language model, DeepSeek V3.1, a development hailed as a breakthrough for China’s efforts to strengthen self-reliance in AI and semiconductors.
Launched last week under the MIT open-source license, V3.1 introduces a hybrid reasoning architecture that allows it to switch between “thinking” and “non-thinking” modes. The model delivers more than 40 percent performance gains in benchmarks such as SWE-bench and Terminal-bench compared with earlier versions.
Boost for Domestic Chipmakers
DeepSeek V3.1 is optimized for UE8M0 FP8, a new data format that works seamlessly with locally designed chips. Analysts say this compatibility could significantly boost China’s homegrown semiconductor industry, which has faced pressure from U.S. restrictions on advanced foreign processors.
The announcement immediately lifted investor sentiment, sending shares of Chinese chipmakers higher. Cambricon Technologies surged nearly 20 percent to record levels, while SMIC and Hua Hong Semiconductor also logged double-digit gains.
Cambricon, a major supplier to DeepSeek, reported a dramatic turnaround in its financials for the first half of 2025, posting a 1.03 billion yuan profit versus a 533 million yuan loss a year earlier. Revenue rose to 2.9 billion yuan, largely attributed to soaring demand for AI chips following DeepSeek’s release.
Strategic Value
Industry experts describe DeepSeek’s V3.1 as a “key pillar” in China’s AI sovereignty drive, allowing developers to train and deploy cutting-edge systems without relying on U.S. technology. The move underscores Beijing’s broader strategy of promoting domestic innovation in response to ongoing trade and tech restrictions.
What’s Next
DeepSeek is expected to deepen its partnerships with Chinese hardware firms while expanding real-world applications of its AI models in areas such as education, finance, and manufacturing. Analysts predict more companies will adopt locally optimized systems, accelerating the country’s transition toward a self-sustaining AI ecosystem.
