← Back to the directory

Company · Works with Zhejiang University, MoXin Technology

Huawei

华为


Coverage11

Release · Oct 8, 2026

Huawei Opens Public Beta for Three HarmonyOS PC Native Development Tools: DevEco Studio, DevEco Code and DevEco CLI

On October 8, Huawei opened a public beta for three HarmonyOS PC native development tools: DevEco Studio, DevEco Code, and DevEco CLI. DevEco Studio is a visual all-in-one IDE built on Huawei's self-developed BitFun (毕方) engine; it supports emulator-free multi-device parallel debugging and intelligent code completion.

DevEco Code is a one-stop AI development tool for HarmonyOS apps, offering end-to-end agentic development with controllable Goal, Plan, and Build layers. DevEco CLI packages the HarmonyOS toolset, an official knowledge base, and over 70 Skills into an AI-callable command-line interface. Huawei says the installed base of HarmonyOS 6 and HarmonyOS 7 hardware has surpassed 90 million units.

Original sources (Chinese)

Agentic开发时代到来!鸿蒙PC三大原生开发工具公测,补齐国产操作系统生态关键短板zhidx

Release · Oct 2, 2026

openJiuwen Debuts X-Router Self-Evolving Model Routing Technology, Ascend-Friendly, Cutting Agent Token Consumption by 50%+

X-Router, a self-evolving model routing technology, has been released by openJiuwen, an open-source AI agent platform co-built with Huawei 2012 Labs, Huawei Cloud, and other Huawei units plus university and enterprise developers. It dynamically selects and orchestrates models based on task complexity, combining profiling, decision, and evolution layers with system states such as KV cache affinity, real-time load, and target availability for multi-objective optimization. It learns from feedback through contextual bandits and lightweight reinforcement learning. The platform says it is Ascend-friendly and reduced token consumption by more than 50% in agent testing.

Original sources (Chinese)

openJiuwen X-Router自演进模型路由技术首发,昇腾亲和,Agent越跑越省,实测减少50+%Token消耗qbitai

Release · Oct 1, 2026

Huawei launches Mate90 series with Kirin 9050 Pro logic-folding chip supporting on-device 30B MoE LLM

Huawei has launched the Mate90 series: Mate90, Mate90 Pro, Mate90 Pro Max, a Collector's Edition and an RS Ultimate Design model, all on Kirin 9030/9035/9050 Pro chips and HarmonyOS 7. The top Kirin 9050 Pro uses a "logic folding" vertical stacked architecture; Huawei says multi-core performance rises 23%, GPU rendering 40%, and NPU 140%, enabling a 30B MoE LLM on-device for offline AI photo editing. Cameras introduce a third-generation red maple primary color sensor (color accuracy up 37%), an 18EV high dynamic range main camera, a 200MP telephoto, and a Ruijing Z10 ten-fold continuous optical zoom module. Mate90 starts at about $857 (RMB 5,999); the Pro Max Collector's Edition camera bundle is about $2,786 (RMB 19,499).

Original sources (Chinese)

顶配18999元!华为新旗舰外挂巨炮相机,余承东解读逻辑折叠韬芯片zhidx

Release · Oct 1, 2026 · as partner

MoXin Technology launches 4D world model MoWorld, debuting on-device via HarmonyOS app

MoXin Technology has released MoWorld, a native 4D world model, now shipping with the Huawei Mate 90 series through the HarmonyOS-exclusive app Touch Pet (摸小宠). Users upload a few pet photos and get a rotatable digital 4D pet with continuous motion.

The device-cloud pipeline uses cloud Ascend NPU diffusion to generate 3D Gaussian assets, then a roughly 600-million-parameter MoWorld-2.0-Flash model renders on the Kirin chip after quantization and operator adaptation. The company reports 3D/4D Gaussian rendering above 500 FPS and around 100 FPS on the phone, with inference cost about 30% of an equivalent imported GPU solution. It has also added physics properties such as gravity and elasticity.

Original sources (Chinese)

00后浙大博士把4D世界模型首次塞进手机!aiera

Open source · Sep 30, 2026 · as partner

DeepSeek open-sources Ascend infrastructure components in collaboration with Huawei Ascend

The Ascend counterpart to DeepSeek's GPU infrastructure stack has been open-sourced, including the TileLang high-level compiler, DeepGEMM, FlashMLA, TileKernel and DeepSelect kernel libraries, and the DeepEP distributed communication library. Huawei contributed the jointly defined SuperPoD Flex and UBL128 fabric with 128-card 3.2 Tbps single-layer scale-up and 256,000-card two-layer scale-out. Reported DeepEP throughput is 375 GB/s for dispatch and 347 GB/s for combine; on Ascend 950, DeepSeek-V4.1-Flash offline inference runs at 2,469 tokens/s per card at TPOT=5ms. Licence terms were not stated.

Original sources (Chinese)

DeepSeek官方开源昇腾基础组件,与昇腾共建高效易用的AI芯片软件生态qbitaiDeepSeek 开源面向华为昇腾算力平台的基础设施组件,与英伟达平台一一对应ithome

Release · Sep 30, 2026

Huawei unveils Hi-ONE NPO optical engine deployment: 5,500 units for 4,096-card Ascend 960 supernode

Huawei has announced a deployment plan for its HiSilicon-developed Hi-ONE optical engine, using near-package optics (NPO) in a 4,096-card Ascend 960 supernode. The design replaces 48,000 800G pluggable optical modules with 5,500 Hi-ONE units, each carrying 7.2T of bandwidth across 36 200G lanes. A cross-cabinet interconnect device supports 176 1.6T ports and about 280T of all-optical interconnect bandwidth. Huawei says this would be the first AI supernode to use NPO.

Huawei also launched an OPEN NPO standards initiative with Baidu, JD Cloud, China Mobile Research Institute and optical component makers.

Original sources (Chinese)

5500颗光引擎如何撑起4096卡超节点?华为NPO的故事还没讲完leiphone

Release · Sep 28, 2026

Huawei launches Luxeed RX with L3 autonomous driving architecture, begins public-road testing

At a Luxeed RX and HarmonyOS Smart Driving launch event on September 28, Huawei executive director Richard Yu introduced the Luxeed RX as a vehicle built on an L3 autonomous driving architecture, saying it has already received an L3 road-test permit and is being tested on urban expressways. A test video shows a steering-wheel “A” button that activates L3 mode, with Xiaoyi announcing the status and the center screen displaying “L3 autonomous driving in progress.” The car also carries the CAS 5.0 all-around anti-collision system, which Huawei describes as six-dimensional safety. Consumer availability of L3 driving functions was not announced.

Original sources (Chinese)

华为余承东:冗余是 L3 时代入场券,也是自动驾驶的双重保险ithome华为余承东:智界 RX 提前迎接 L3 自动驾驶时代,正在公开城市快速路等复杂道路开展实测ithome

Partnership · Sep 20, 2026 · as partner

Nanfang Hospital and Huawei deploy HAIP platform, enabling clinicians to build medical AI agents without coding

Nanfang Hospital of Southern Medical University and Huawei have deployed the HAIP platform on Huawei's DCS AI solution. Its Nexent agent platform lets clinicians describe clinical logic in natural language to generate agent prototypes without coding; nephrology staff used it to build a multi-specialty screening agent for rare disease aHUS, going from zero experience to a preliminary prototype in ten days. Obstetrics and the liver disease center are also using it. No contract value or timeline was disclosed.

Original sources (Chinese)

在南方医院,临床医生开始「造」AI工具leiphone

Open source · Sep 20, 2026 · as partner

China Telecom AI open-sources Xing4.0-29B-A4B full-stack domestic code agent LLM

Xing4.0-29B-A4B, a mixture-of-experts code agent model with 29 billion total parameters and about 4 billion active, has been open-sourced by China Telecom AI. At 4-bit quantization it fits on a single RTX 3090, and it was trained on Huawei Ascend 910C using MindSpore and MindFormers. The model is available on GitHub, Hugging Face, Gitee, ModelScope and other hubs with an API, and in a private customer-service deployment China Telecom AI reports a complex business task success rate above 90%.

Original sources (Chinese)

一张3090就能跑!全栈国产模型,把AI办公搬到企业本地qbitai

Release · Sep 4, 2026 · as partner

Qijing GX7 officially opens pre-sale with L3-level autonomous driving architecture

Pre-sales opened September 4 for the Qijing GX7 (启境GX7), a wide five-seat SUV jointly developed by GAC Group and Huawei under the Qijing Motors brand. Three range-extended versions are offered — Ultra, Ultra dual-motor and Ultra triple-motor — priced from $35,700 to $44,600 (RMB 249,900 to 311,900). The GX7 carries Huawei Qiankun ADS 5 and what the partners describe as an L3 autonomous driving architecture, along with CATL's Xiaoyao battery. The vehicle is scheduled to go on sale officially within September.

Original sources (Chinese)

起步即搭载L3架构,阔五座开创者启境GX7正式预售24.99-31.19万leiphone

Release · Aug 16, 2026 · as partner

openJiuwen releases WorkSwarm Swarm Office Agent, supporting HarmonyOS PC and more platforms

WorkSwarm (蜂群办公智能体), a multi-agent office tool from openJiuwen, has debuted on the HarmonyOS PC app market and also supports Windows and Mac. The company says its agents can autonomously form teams, orchestrate tasks, share context, and pass results to one another to handle office work, programming, and music creation, with humans able to join the collaboration. Pricing and performance benchmarks were not disclosed.

Original sources (Chinese)

WorkSwarm:引领办公智能体新范式,让AI从一个助手,进化为一支与你并肩作战的团队qbitai