Huawei co-produces safety -focused DeepSeek model with 1000 AI chips
During the Connect 2025 conference, Huawei mentioned how its Ascend AI chips will develop a new and improved DeepSeek AI model for some “safety-focused” purposes. Now this creation is coming into light, revealing its tech-pack capabilities.
According to the details, Huawei and a Chinese University have jointly developed a new and safety-focused DeepSeek AI model using 1000 Ascend chips. This creation can successfully prevent discussion of sensitive topics or political subjects up to 100%.
Chinese regulators will primarily use this model to keep an eye on socialist values with tight control over speech. It will avoid around 14 malicious threats and risk factors.
Named DeepSeek-R1-Safe, the model will keep an eye on toxic and harmful speech, politically sensitive content, and illegal behavior that could distress the country.
Huawei has co-developed the new DeepSeek-R1-Safe AI model with China’s Zhejiang University. It is an improved version of the original one and can outperform Alibaba Cloud’s Qwen in producing results, complied with China’s rule book.
Huawei, said:
Its comprehensive security defence capability reached 83%, outperforming multiple concurrent models like Qwen-235B and DeepSeek-R1-671B by 8% to 15% under identical testing conditions.
“DeepSeek-R1-Safe exhibited a less than 1% performance degradation compared to the original DeepSeek-R1.”
The report hints that DeepSeek is not involved in this development. Also, the Chinese startup hasn’t responded to this matter yet. Though the new model seems to be a helpful product for the Chinese government to maintain harmony in the country.
Other features and changes of the DeepSeek-R1-Safe are currently unknown. Although the technology will soon be available for commercial use. Stay tuned.
READ the latest news shaping the AI Chips market at AI Chips News
Huawei co-produces safety -focused DeepSeek model with 1000 AI chips, source
Add comment