At the end of July, a new type of certificate marked with L3 began circulating among the marketing teams of Chinese smartphone manufacturers. This mark is based on a national standard published by China's Ministry of Industry and Information Technology in May for artificial intelligence-enabled terminals.
The first batch of test results, released on July 17, included 11 qualified devices. These comprised nine smartphones and two tablets from brands such as Huawei, Motorola, Honor, vivo, OPPO, Xiaomi, and Stepfun.
It is important to note that this certification is not a regulatory barrier; a phone does not need to pass any tests to be released with the L3 mark, nor did the models not included in the first batch fail any tests. The standard serves as a guideline for carriers, application developers, and procurement groups, helping to determine which devices can perform planning tasks, make calls, and remember user preferences, versus those that remain merely voice interfaces of chatbots.
The L3 marking is also not the pinnacle of the hierarchy. The MIIT document defines five dimensions—perception, cognition, execution, memory, and learning—as well as 14 sub-capabilities with clear levels from L1 to L4. However, the initial round of testing only covered level L3. Nevertheless, manufacturers are actively using this terminology, referring to L3 as the current highest level.
Technically, this description is correct, but it omits the second half: level L4 has been intentionally left blank because the rules concerning delegation between agents, cross-device collaboration, and error attribution have not yet been developed. This distinguishes phone agents from autonomous driving systems, despite the common L1 to L4 scale. Automotive ratings revolve around the transfer of responsibility between the driver and the vehicle, whereas phone ratings focus on task complexity, tool orchestration, and memory.
A phone capable of planning a multi-day trip, including booking hotels, calendars, and payments, must still hand over control of any irreversible actions to the user. Thus, the empty space at L4 indicates not an incomplete model, but that interaction rules are not yet ready.
For end consumers, the situation is simple. Most flagship phones already meet Level L2, which functions as a chatbot with good tool-calling capabilities. L3 devices are already available for purchase, such as the Huawei Pura X Max and the Motorola line from Lenovo. Furthermore, using L3 features does not require purchasing a new gadget. The next significant step, when it comes, will be related not to hardware upgrades, but to the adoption of new regulations.