AI News Feed
Market watch
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
Companies

OPPO and OpenKG Launch MobileMem, Industry's First Benchmark for On-Device AI Memory

OPPO and OpenKG have launched MobileMem, the first benchmark for evaluating on-device AI memory, to standardize testing and accelerate industry development.

AI News Feed · 2026-09-02 15:19

The benchmark addresses a gap in systematic evaluation for mobile-side long-term personalization and multimodal life scenarios. Unlike cloud-based agents, on-device AI must accumulate, update and use long-term user memory under constraints of limited computing power, storage and privacy protection. Without an open reference, developers of edge-side agents have struggled to run effective A/B tests and performance regression, while researchers lacked consistent tasks and evaluation protocols to compare different solutions.

MobileMem's evaluation dataset contains historical contexts of up to one million tokens to simulate long-term real-user interactions. It focuses on three core technical dimensions: modeling and retrieval of long-term events across time, applications and modalities; cross-modal association and reasoning between text and images; and continuous evolution and personalization adaptation, including updates to user preferences, social relations and event states.

The benchmark covers scenarios such as a personal life companion. For example, when a user asks for a vacation trip similar to one enjoyed last year, the assistant should draw on long-term memory to generate a personalized itinerary reflecting historical preferences, travel habits and spending patterns, while avoiding previously encountered inconveniences.

OPPO and partners say the test framework is compatible with multiple agent systems, covering task understanding, memory retrieval, search and execution modules. Jiang Yuchen, OPPO's director of ColorOS smart product R&D, said: "Long-term memory is an important foundation for on-device AI to deliver personalized intelligence. MobileMem is not just a benchmark but industry infrastructure to accelerate the landing of memory technology."

Wang Haofen, rotating chair of OpenKG and a professor at Tongji University, said MobileMem provides a unified reference for comparing and iterating different technical routes, promoting open academia-industry research collaboration. Chen Huajun, a professor at Zhejiang University, said future edge-side agents need long-term memory, knowledge accumulation and continuous learning, and a systematic evaluation system is essential for advancing the field.

The MobileMem 1.0 white paper is now open to the industry, and the evaluation datasets, task sets and related tools will be progressively open-sourced. OPPO said it welcomes industrial and academic participation in the ongoing development of MobileMem.