Kunlunxin Gains Early Edge in AI by Porting Model on Release

According to AI TIMES, Kunlunxin completed its "Day 0" adaptation to verify and deploy MiniCPM5-2B on its platform immediately after its release on September 15, securing an environment where it can run at once on domestic AI semiconductors. The model is a small edge model that supports a 512K long-context window, and through its partnership with FlagOS, it has built an expansion base aimed at smartphones, PCs, and in-vehicle devices. The key point is not the model's performance itself, but that the company enabled a new open-source model to be moved into a real computing environment as soon as it appeared. Because each semiconductor has different computing methods and execution environments, separate optimization is required, and this case is meaningful in that it reduced that process into a common software layer. Based on this, Kunlunxin is broadening support for other open-source models and strengthening an ecosystem-linking strategy beyond responding to specific models. The article notes that competition in generative AI is shifting from chip performance alone to how quickly models, software, and semiconductors can be tied together.