On August 24, 2026, Xiaomi held a Xuanjie Chip Technology Briefing in Beijing, unveiling three self-developed chips: the Xuanjie O3 high-end flagship SoC, built on a 3nm process with 24 billion transistors and an AnTuTu score of 5.22 million; the Xuanjie O100, a 6nm edge-side AI accelerator chip with a bandwidth of 1.22 TB/s; and the Xuanjie D100, a 3nm intelligent driving chip. The launch of these three chips marks Xuanjie’s expansion beyond mobile chip business, establishing it as the AI computing foundation supporting Xiaomi’s full ecosystem of people, cars, and smart homes, enabling seamless coverage from pocket to cockpit.Article author and source: AI World
On August 24, 2026, Xiaomi held the Xuanjie Chip Technology Briefing in Beijing, officially launching three Xuanjie chips: the Xuanjie O3, a cutting-edge AI flagship SoC; the Xuanjie O100, an AI accelerator chip with 1.22 TB/s high bandwidth; and the Xuanjie D100, a high-performance AI chip for intelligent driving. The simultaneous release of these three chips represents a significant expansion of Xiaomi’s chip business, marking Xuanjie’s transition beyond single-device smartphone chip design to meet the demands of AI technological advancement and diverse application scenarios, becoming the foundational AI computing platform supporting Xiaomi’s full ecosystem of people, cars, and smart homes.
Xuanjie O3: Xiaomi's next-generation high-end flagship SoC, fully embracing AI
Xuanjie O3 is Xiaomi's second self-developed high-end flagship SoC, continuing the 3nm process technology while increasing the transistor count from 19 billion in the previous generation to 24 billion, delivering powerful performance. According to Xiaomi Lab data, the Xuanjie O3 achieves an AnTuTu benchmark score of 5,228,014, making it the first mobile flagship platform in the industry to surpass 5 million points.
In terms of CPU, the Xuanjie O3 features a 10-core all-big-core architecture, combining six ultra-big cores and four big cores with a maximum clock speed of up to 4.35 GHz, delivering leading baseline performance on mobile devices. It achieves a multi-core GeekBench 6 score of up to 15,221, representing a significant 60% improvement over the previous generation. In terms of energy efficiency, the Xuanjie O3 continues to excel in mid-to-low load scenarios, reducing power consumption by an additional 25% in everyday usage.
The graphical capabilities of Xuanjie O3 have also undergone a major upgrade, featuring the newly launched G2-Ultra NX graphics processor and continuing with a 16-core ultra-large-scale configuration. Traditional graphics performance has increased by 85% compared to the O1, while ray tracing performance has improved by 182%. Additionally, the Xuanjie O3 has been significantly optimized for GPU energy efficiency, achieving an energy efficiency leap across all scenarios—with power consumption reduced by up to 64% compared to the previous generation at equivalent performance levels.
In addition to its consistently powerful computing units, the Xiaomi Xuanjie team has significantly enhanced storage-related configurations to accelerate data delivery and improve the chip’s underlying operational efficiency. The Xiaomi Xuanjie O3 features a premium on-chip cache totaling 60MB, including a new 16MB SLC cache that addresses the shortcomings of the previous generation. It is the first to support LPDDR6 memory, delivering a memory bandwidth of up to 113.8 GB/s—a 48% improvement. Xiaomi has also innovatively implemented the Xuanjie Unified Fusion Bus Architecture, reducing memory access latency to just 82ns, setting a new industry benchmark.
The Xuanjie O3 also features a premium multimedia module, with Xiaomi's fifth-generation ISP receiving comprehensive upgrades: it supports up to 432 megapixel single cameras, offers a maximum internal image signal processing bit depth of 24-bit, and enhances AI RAW-domain noise reduction for night video to 4K/60fps. The DPU expands screen分区 tone mapping to more scenarios, improving screen readability in low light and enhancing HDR content display. The VPU adopts a separated encoding/decoding architecture, boosting video editing export speeds by 20%, and introduces the industry’s first H.266 hardware decoding on Android.
In addition, the Xuanjie O3 features a fully reengineered security module, achieving national cryptographic and CCRC EAL5+ security certifications. On the baseband side, its communication performance matches that of leading flagship platform integration solutions, with overall power consumption in 5G scenarios reduced by more than 20% compared to the previous generation.
Xuanjie O3 is not only a smartphone SoC that delivers a cutting-edge contemporary experience, but also an AI-focused SoC fully embracing the AI era. AI hardware units are integrated within all primary cores to meet light AI algorithm demands across various scenarios, reducing the need to frequently invoke the NPU and thereby minimizing latency and additional power consumption caused by inter-module communication within the chip.
The NPU has undergone extensive hardware-level optimizations for large language models, featuring a quad-core architecture that delivers 200 TOPS of computing power, significantly surpassing industry-leading flagship solutions. The Xuanjie O3 also introduces the industry’s first mobile SIMT computing architecture with a Vector unit, providing a leading 3.13 TFLOPS of vector computing power to overcome hardware bottlenecks for on-device large model execution. Additionally, the Xuanjie O3 NPU has increased near-memory investment, combining 5-value quantization developed with Xiaomi MiMo and hardware-assisted Huffman lossless compression to achieve a 45% faster model inference speed compared to mainstream flagship SoCs.
Xuanjie O100: 1.22 TB/s high-bandwidth AI acceleration chip, strengthening the hardware foundation for the Agent era
Xuanjie O100 is a 6nm edge-side AI acceleration chip independently developed and designed by Xiaomi. When paired with Xuanjie O3, it enables ultra-fast inference output for large models on the edge, allowing devices equipped with this platform to quickly respond to user AI requests even in weak or no network conditions.
This chip employs the industry’s most advanced 3D Wafer-on-Wafer stacking technology, using state-of-the-art Hybrid Bonding to thermally bond two layers of DRAM wafers with one layer of NPU compute wafer. The Xuanjie O100 contains 2.58 million bonding nodes with a pitch of just 1.4μm—a specification that surpasses the HBM memory used in cloud AI chips.
Ultra-high-density interconnects between the internal computing and memory units enable the Xuanjie O100 to achieve 16 times the memory bandwidth of mainstream smartphones, enabling on-device large model inference speeds of up to 330 tokens per second, completely resolving the bandwidth bottleneck in on-device large model computation.
The Xuanjie O100 also features a 14-core NPU specifically designed for large models, paired with Xiaomi’s innovative XRING HB-Matrix high-bandwidth matrix bus, enabling efficient collaboration among internal cores and rapid completion of AI computation tasks.
In addition, Xiaomi addressed numerous challenges in the chip manufacturing process through innovative design, such as spending six months iterating the chip layout to solve warping and cracking issues during high-temperature wafer bonding. The three-layer chip structure within the Xuanjie O100 features an F2F metal layer direct-connect architecture, for which multiple patents have been granted.
Xuanjie D100: High-performance AI chip for autonomous driving, completing the full ecosystem integration of person, vehicle, and home.
Xuanjie D100 is Xiaomi's self-developed 3nm high-performance AI chip for intelligent driving, and it is also China's first 3nm intelligent driving chip, featuring 20 high-performance CPU cores and a powerful 16-core high-compute NPU.
In addition to efficiently running autonomous driving algorithms in the automotive domain, the Xuanjie D100 also delivers ultra-high AI computing power locally for individual users, supporting up to 160GB of memory per chip and enabling the local deployment of large models with over 200 billion parameters. This provides professional developers and tech enthusiasts with local AI inference capabilities, eliminating costly cloud API usage while physically eliminating the risk of privacy data leakage.
In addition, the Xuanjie D100 enables multi-chip integrated computing through the Xuanjie High-Speed Interconnect Bus, delivering greater hardware computing power to handle complex AI tasks locally.
These are the three new chips launched by Xiaomi Xuanjie: the Xuanjie O3 will debut this September alongside the Xiaomi 18 Fold, while the Xuanjie O100 and D100 have completed development and will be commercially available next year.
The launch of these three Xuanjie chips signifies that Xuanjie has become Xiaomi’s full-ecosystem AI computing foundation, spanning from pocket to cockpit, from living room to factory—unifying all human-vehicle-home scenarios with a single Xuanjie hardware solution to meet high-efficiency, high-bandwidth, and high-compute AI demands across diverse contexts. It also marks Xiaomi’s steady and confident leap toward becoming a leading hard-tech innovator, powered by its strengthened chip design capabilities.