On February 5th, 2025, Lenovo's personal AI assistant "Xiao Tian" integrated the highly talked-about DeepSeek AI large model. Users can now access DeepSeek through Xiao Tian's AI Space interface. At the same time, Huawei's system-level AI "Xiao Yi" also added the new DeepSeek R1 model on HarmonyOS NEXT (the native version of HarmonyOS), marking the launch of DeepSeek R1 with the Xiao Yi app upgrade (version 11.2.10.310).
The Strategic Significance of DeepSeek-R1 Integration
The simultaneous integration of DeepSeek by Huawei and Lenovo is not a coincidence. While Huawei is positioning its system as a core part of the "full-scenario smart life" within the HarmonyOS ecosystem, Lenovo's approach allows for cross-terminal access to personal smart devices. The rapid popularity of DeepSeek-R1 is attributed to two key factors: efficiency and open-source access. DeepSeek-R1 offers performance similar to OpenAI's models but at a significantly lower cost, making it accessible to a wider range of manufacturers and users, and its open-source nature allows companies to freely integrate and optimize the model.
In just two weeks, DeepSeek-R1 has quickly gained recognition in both the tech industry and consumer market, becoming a go-to AI solution for a range of companies including Keep, Huawei, and Lenovo.
Challenges in Integration and Experience
Despite the widespread enthusiasm, the integration of DeepSeek-R1 into Huawei and Lenovo's systems is still far from perfect. For instance, Huawei's approach to integrating DeepSeek R1 is more of a plugin than a deeply embedded feature. In HarmonyOS NEXT, DeepSeek-R1 appears as an external "intelligent agent," which users can access via the Xiao Yi app. However, DeepSeek R1 does not serve as the default AI engine in the system, meaning the regular user experience remains unchanged unless explicitly activated. This "plugin-style enhancement" offers new AI capabilities but does not overhaul the underlying system architecture.
Huawei's hesitation to fully integrate DeepSeek R1 could be due to the company’s ongoing development of its own AI model, the Pangu model, which competes with DeepSeek. Additionally, the full version of DeepSeek-R1 requires significant computational resources, and local deployment on personal devices remains a challenge for manufacturers. With versions of DeepSeek-R1 ranging from 1.5B to 70B parameters, it may not be optimized for personal devices without further refinement.
