--- title: "China grades its AI gadgets: a new standard sorts phones, TVs, and glasses from L1 to L4" date: 2026-11-01 category: Policy & Governance site: NeuroAI canonical: https://neuroai.site/a/na-policy-ai-terminal-grading language: en --- # China grades its AI gadgets: a new standard sorts phones, TVs, and glasses from L1 to L4 > Released on 8 May 2026 by MIIT, SAMR, and the Ministry of Commerce, the GB/Z 177—2026 series grades AI terminals on a four-level ladder from L1 response to L4 coordination. You are shopping for a new phone. Two models both call themselves "AI." One nudges you with reminders; the other plans your day, talks to your car, and learns your habits. Until now, nothing told you which was actually smarter. A new Chinese national standard tries to fix that. ## Key takeaways - The "AI Terminal Intelligence Grading" (人工智能终端智能化分级) series, GB/Z 177—2026, was released on 8 May 2026 by the Ministry of Industry and Information Technology, the State Administration for Market Regulation, the Ministry of Commerce, and other departments. - It uses a "2+N" architecture: two foundation standards (reference framework plus general requirements) and a set of product-specific standards. - Devices are graded on a four-step ladder, L1 (response) to L4 (coordination), with the top level, L4, to be refined in a later revision. - The first batch covers seven product categories: mobile terminals, microcomputers, TVs, glasses, car cabins, speakers, and headphones. - GB/Z is a recommendatory guidance technical document (国家标准化指导性技术文件), not a mandatory national standard — a distinction that matters for what it can compel. ## The "2+N" architecture, explained The series is built in two layers. The "2" is a pair of foundation standards — Part 1, the reference framework, and Part 2, general requirements — that define what an AI terminal is, how to grade it, and how to test it. These answer the basic questions every product category will lean on. The "N" is the set of product-specific standards that hang off those foundations: phones, computers, TVs, glasses, car cabins, speakers, headphones, and more. The first batch names seven categories, with others to follow. ## The L1 to L4 ladder The grading system runs from the most passive to the most autonomous: - **L1 — response level** (响应级): the device reacts to commands. - **L2 — tool level** (工具级): it performs defined tasks as a smarter tool. - **L3 — assistant level** (辅助级): it understands intent and assists proactively. - **L4 — coordination level** (协同级): it coordinates across tasks and contexts. The top rung, L4, is explicitly flagged as something to be clarified and improved in a future revision as the industry matures. So for now, the practical ceiling most products are measured against is L3. Read as a roadmap, the ladder also tells makers what "better" means at each step: L1 is reactive hardware, L2 is a smarter tool, L3 is a device that understands intent and acts on it, and L4 is one that coordinates across the user's other devices and tasks. A phone maker does not have to guess what capability separates its flagship from a rival's; the standard names it. That clarity is the quiet utility of the whole exercise — it turns "AI" from a marketing adjective into a testable claim. ## Which products come first The first batch deliberately covers the devices people already own or are buying fastest: mobile terminals, microcomputers, TVs, glasses, car cabins, speakers, and headphones. According to testing and analysis cited by state media from the China Academy of Information and Communications Technology (CAICT, 中国信息通信研究院), most widely held products today sit at L1 or L2, with some newer releases reaching L3. The same source noted that AI phones, PCs, and tablets have already shipped in the tens of millions, making them the current volume leaders. That spread is the point. A label telling a buyer "this is L2, that is L3" turns a marketing word into a comparable spec. ## What it means for buyers and makers For shoppers, the grading system is meant to become a shortcut: a way to compare devices on intelligence rather than on spec-sheet noise. For makers, it sets a common test method, which lowers the cost of proving a product's capability to regulators, retailers, and export markets. The standard is also designed to plug into consumer policy. The releasing bodies said the grading should be applied inside the 2026 consumer trade-in (以旧换新) framework, helping build a catalog of AI terminal products to guide public purchasing decisions. According to Li An, chief scientist at BrainNet (脑机网), the bigger effect may be export-related: a recognized national grading scheme gives Chinese device makers a ready-made vocabulary to defend their products against foreign "AI capability" labeling rules. ## Honest limitations This article is based on state-media reporting (CCTV and Economic Daily) of the standard's release and the accompanying explanatory briefing; we did not retrieve the full GB/Z 177—2026 text from the national standards platform to verify every clause. The "tens of millions" shipment figure and the L1/L2/L3 distribution claim come from a CAICT engineer quoted by state broadcasters, not from an independently published dataset, so treat them as indicative rather than audited. Because GB/Z is a recommendatory guidance document, not a mandatory standard, its enforcement power depends on how regulators and retailers choose to adopt it. We have not confirmed a specific implementation date beyond the 8 May 2026 release. ## What readers can do now - If you are buying an AI device in China, start checking for an intelligence-grade label (L1–L4) before trusting a vendor's "AI" claim at face value. - If you make or import consumer electronics, map your products to the L1–L4 ladder now; the grading is being tied to the 2026 trade-in catalog. - If you follow standards, watch for the "N" product-specific standards beyond the first seven categories — that expansion is where compliance work will actually land. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-ai-terminal-grading Free to quote with attribution and a link to the original.