Xiaomi's Xring O3 Breaks 5 Million on AnTuTu — and Unveils a Three-Chip Silicon Strategy

Xiaomi has been designing its own silicon since the Xring O1 in 2025. On August 24, 2026, at its Xring Chip Technology Communication Conference in Beijing, the company argued it has graduated from follower to contender — unveiling not one flagship chip but a three-chip portfolio that spans phones, on-device AI, and autonomous driving, all fabbed by TSMC.

The headline: the Xring O3, a 3nm flagship SoC that Xiaomi says scored 5,228,014 on AnTuTu v11 — the first mobile processor to clear five million points. But the more consequential announcements may be the two chips flanking it: the Xring O100, a 6nm neural processing unit with 1.22 terabits per second of near-memory bandwidth built to run Xiaomi’s MiMo large language model entirely on-device, and the Xring D100, described as China’s first homegrown 3nm intelligent-driving processor, capable of running models with up to 200 billion parameters locally.

Together they sketch a clear strategy: vertical integration of hardware and AI that reduces reliance on Qualcomm and MediaTek while tying Xiaomi’s phones, electric vehicles, and robots into one silicon stack.

The Xring O3: all big cores, no apologies

The record AnTuTu score is a by-product of a genuinely unusual design decision. Every other Android chipmaker — Qualcomm with the Snapdragon 8 Elite, MediaTek with the Dimensity 9400, and Xiaomi’s own Xring O1 — divides CPU cores into performance and efficiency tiers. The O3 discards the efficiency tier entirely: a ten-core, all-big-core CPU split into two C1-Ultra cores at up to 4.35 GHz, four C1-Premium cores at 3.68 GHz, and four C1-Pro cores at 3.15 GHz.

Xiaomi reports 60% better multi-core performance than its baseline, with a Geekbench 6.5 multi-core score of 15,221 — a 61% jump over the O1. The trade-off is real: all-big-core designs optimize for sustained workloads at the potential cost of idle efficiency, and whether Xiaomi’s power management can compensate is a question only retail hardware will answer.

The rest of the spec sheet:

Spec	Xring O3
Process	TSMC 3nm (N3P), 24B transistors, 133 mm² die
CPU	10-core all-big-core (2× 4.35 GHz + 4× 3.68 GHz + 4× 3.15 GHz)
GPU	16-core in-house G2-Ultra NX, +85% perf, +64% efficiency
Memory	LPDDR6 at 10,667 Mbps → 113.8 GB/s (+48% vs O1)
NPU	200 TOPS tensor, 3.13 TFLOPS vector (+45% vs O1)
AnTuTu v11	5,228,014 (claimed, lab conditions)

The LPDDR6 support is a genuine first for a shipping mobile SoC. JEDEC finalized the standard (JESD209-6) in July 2025, and LPDDR6 changes memory architecture structurally — from two 32-bit sub-channels to four 24-bit sub-channels — prioritizing concurrency and latency over raw clock speed. That is precisely the profile on-device AI inference needs, where the bottleneck is repeatedly hauling model weights between memory and compute, not arithmetic throughput.

On AI, Xiaomi also diverges from the single-NPU orthodoxy: acceleration is distributed across the CPU, GPU, ISP, DPU, and ADSP, and the company claims 82-nanosecond static memory latency — a figure it says beats Apple’s A19 Pro on that specific measurement, though it awaits independent validation.

Development reportedly took just 459 days. Reuters, citing sources, reported the chip has entered mass production with a shipment target of 200,000–300,000 units, at an estimated average price of 1,329 yuan (about $198).

Debut devices: Xiaomi 18 Fold and Pad 9 Pro Max

The O3 will first appear in the Xiaomi 18 Fold and Pad 9 Pro Max, both launching in China in September. Preorders for the 18 Fold are already open, though pricing is unannounced and the device may well remain China-only, like its MIX Fold predecessors.

The competitive context is stark: one research firm puts Huawei’s share of China’s foldable market at roughly 68% in Q2 2026, with 1.6 million units shipped. Xiaomi’s 200–300k target is a fraction of that — but it follows an O1 generation that shipped in three devices and cleared one million units combined. The cadence is real; the scale is still being built.

The O100: attacking the memory wall

Arguably the most architecturally significant chip of the three isn’t the one with the benchmark record. The Xring O100 is a 6nm inference accelerator whose defining feature is a near-memory architecture — stacking compute and memory vertically, connected by millions of high-speed channels, to collapse the energy cost of moving data between them.

This targets the “memory wall” directly. Research published in IEEE Computer in May 2026 by Google DeepMind scientists identified data movement — not compute throughput — as the primary bottleneck in LLM inference, and Fudan University work has estimated data-movement energy at roughly two orders of magnitude above computation energy in conventional designs. Xiaomi claims the O100 delivers 1.22 Tbps of memory bandwidth, roughly ten times the O3’s figure — enough to run the MiMo LLM on smartphones, vehicles, and robots without a cloud connection. Qualcomm’s own High-Bandwidth Compute near-memory announcement in June validates the direction independently. The O100 has completed development and is targeted for 2027 deployment.

The D100: 200B parameters in a car

The Xring D100 extends the portfolio into automotive: a 3nm chip with a 20-core CPU, 16-core NPU, and support for up to 160 GB of unified memory. Xiaomi claims it can run AI models with up to 200 billion parameters locally — frontier-scale reasoning inside a vehicle, with no cloud dependency. It has completed validation and is planned for commercial deployment in 2027, aligned with Xiaomi’s EV expansion.

The caveats worth keeping

Two sober notes temper the launch. First, the process-node gap: Apple, Qualcomm, and MediaTek are all preparing flagship processors on TSMC’s 2nm process for late 2026, while the O3 ships on 3nm — and node leadership typically decides who wins the sustained-performance battle inside a thin chassis. Xiaomi’s answer is implicitly that its primary battleground is China’s premium market, where the comparison set differs.

Second, the market itself: IDC projects global smartphone shipments to fall roughly 13% in 2026, compressing the volume Xiaomi needs to amortize its chip investments — pressure already visible in Xiaomi raising prices on 11 phones and tablets as memory costs bite.

Neither caveat dims the core achievement. A phone maker that first taped out a flagship SoC barely a year ago has now shipped a three-chip family covering the three compute domains — pocket, edge AI, and vehicle — that will define the next decade of consumer hardware. The Xring O5 or O6, not the O3, will show whether Xiaomi can be genuinely competitive at global scale. But the company has made clear it intends to still be at the table when that question is answered.