EastVoice · AI · 2026-10-22 · Issue #105

Editor's Note: China's AI news on Wednesday was about plumbing and balance sheets. OpenAI and PerturbAI ran close to 2,000 agents in parallel over biology data, Alibaba, Tencent and Baidu together spent ¥131.85 billion on capital projects in a single quarter, and Tencent Cloud put a hundred-billion-scale search index behind an API at ¥14 per thousand calls. On the hardware side, Unitree showed an 8-hour bodyguard robot, Chery's Moja priced a home companion at consumer scale, and Momenta put the L4 scale inflection in 2028.

Commentary: The number to watch is not a benchmark score. It is how much of this quarter's spending turns into revenue that does not depend on the next funding round.

---

I. AI Foundation Models

1. OpenAI and PerturbAI Turn ~2,000 Agents Loose on Biology Data

• Summary: PerturbAI said Wednesday that with OpenAI and Scripps Research it built a reasoning framework that runs close to 2,000 AI agents in parallel to read experimental data, search the literature and propose hypotheses for the lab to test.
• Source: ITHome · 2026-10-21
• Editor's Take: The pitch here is throughput, not accuracy. Whether the hypotheses survive a wet lab is a question the announcement does not answer.

2. Alibaba, Tencent and Baidu Spent ¥131.85 Billion in a Single Quarter

• Summary: TMTPost puts combined capital spending at the three Chinese cloud and model giants at ¥131.85 billion in one quarter, up 114% year on year. Alibaba alone accounted for ¥67.68 billion, up 75%.
• Source: TMTPost · 2026-10-21
• Editor's Take: Capex is now the clearest statement of intent in Chinese AI. None of the three is waiting for demand to be proven before building.

3. Tencent Cloud Opens Yuanbao Search at ¥14 per Thousand Calls

• Summary: Tencent Cloud opened Yuanbao Search to outside developers, a stack it says runs on an index in the hundred-billion range, priced from ¥14 per 1,000 calls.
• Source: ITHome · 2026-10-21
• Editor's Take: Metered search is how a model feature becomes an infrastructure line item. Tencent is selling the dependency, not the assistant.

4. SenseTime's Raccoon App Becomes the Company's Front Line

• Summary: Huxiu looks at how the Xiaohuanxiong app has become the most visible product SenseTime has, in a year when alumni spinouts were the company's better-known calling card.
• Source: Huxiu · 2026-10-21
• Editor's Take: A model lab's brand is only as good as its last shipped product. A consumer app with real daily users beats a research citation.

5. MiniMax's M3.1 Flash Takes an Agent Benchmark

• Summary: Leiphone put MiniMax M3.1 Flash through its own tests after the model took first place on an agentic task benchmark, positioning it against the rest of the Chinese flagship field.
• Source: Leiphone · 2026-10-21
• Editor's Take: Benchmark leadership rotates monthly in China. What holds value is whether the model survives the long-horizon tasks agents actually run.

6. iFlytek's AI Promises, Measured Against Delivery

• Summary: TMTPost works through iFlytek's claims against its own numbers, after chairman Liu Qingfeng argued the market undervalues the company and its domestic-chip route.
• Source: TMTPost · 2026-10-21
• Editor's Take: A domestic-chip strategy is a long position with no hedge. The market prices that risk below the story, and it is not obviously wrong.

7. Beijing Stands Up a State Data Group and Drafts a Manufacturing AI Policy

• Summary: At a State Council press briefing, MIIT said it is drafting policies for a new generation of smart manufacturing built around AI. A new centrally owned national data group was announced at the same time, with its chairman and general manager named.
• Source: TMTPost · 2026-10-21
• Editor's Take: Data is where the state is consolidating first. Ownership gets settled before the market is allowed to bid on it.

8. HKUST Open-Sources an "AI University" for Agents and People

• Summary: A team at HKUST open-sourced what it calls a fourth kind of university, a venue where communities of agents and ordinary users build things together.
• Source: AI Era · 2026-10-21
• Editor's Take: Open-sourcing the venue instead of the model is a cheap way to get other people to do the interesting work.

---

II. AI Software & Applications

1. ByteDance's Quietest Earner Is Not Doubao

• Summary: Sensor Tower data cited by Huxiu points to Catbox, not Doubao, as the app bringing in the most revenue inside ByteDance's AI portfolio. User scale and revenue scale are pulling apart.
• Source: Huxiu · 2026-10-21
• Editor's Take: Companion apps have paying users; assistants mostly have free ones. ByteDance now owns one of each and knows which is which.

2. Doubao Phone Assistant Blocks Financial Apps by Default

• Summary: ByteDance updated the blocked-app list for the Doubao Phone Assistant, cutting off banking, trusts, payments, internet finance and credit. The change follows the screen-automation declaration the assistant published for agents.
• Source: ITHome · 2026-10-21
• Editor's Take: An assistant that can move money is a liability, not a feature. Blocking the category is the cheapest way to stay shipped.

3. AI Office Tools Reach 142 Million Users, Tencent Leads

• Summary: QuestMobile's third-quarter report puts combined monthly users for AI office products at 142 million, up 50.6% year on year. Tencent WorkBuddy leads at 32.35 million, ahead of Baidu's Dazi at 9.77 million.
• Source: TMTPost · 2026-10-21
• Editor's Take: A three-to-one lead in a category this young is Tencent's to lose, and it will be defended with distribution rather than model quality.

4. Alibaba Buys Compute, JD Buys Robots

• Summary: TMTPost contrasts the two plans. Alibaba keeps funding data centres inside a three-year cloud and AI programme its CEO says will overshoot the ¥380 billion headline. JD has committed ¥10 billion to robotics by 2028 and 80 related sites.
• Source: TMTPost · 2026-10-21
• Editor's Take: Both are infrastructure bets. One buys accelerators, the other buys floorspace and arms.

5. Chinese Brokers Move Off Their Own AI Platforms

• Summary: More than ten mainstream brokerages have connected to third-party AI platforms including Kimi, Tencent WorkBuddy, Alibaba Qwen and Volcano Engine, shifting from in-house builds to outside ecosystems.
• Source: Huxiu · 2026-10-21
• Editor's Take: Finance was the vertical everyone expected to build in-house. Renting got there first, and compliance never was the moat.

6. A Tsinghua AIR Spin-off Puts Multi-Agent Doctors On Shift

• Summary: QbitAI profiles a multi-agent medical system from a company incubated at Tsinghua AIR, in which an assistant takes patient history and symptoms before a clinician picks up the case.
• Source: QbitAI · 2026-10-21
• Editor's Take: Intake is the part of care that was already a checklist. Automating the form is less glamorous than diagnosis and far easier to deploy.

7. Doubao Starts Selling Inside the Chat

• Summary: Ahead of Double 11, Doubao added Douyin live-commerce rooms to its product recommendations so users can order without leaving the assistant, and began pushing shopping coupons to users.
• Source: Huxiu · 2026-10-21
• Editor's Take: The assistant is turning into a storefront with a chat box attached. That answers a monetisation question, not a product one.

8. A ¥10 Billion AI Compute Order Against ¥300 Million of Cash

• Summary: TMTPost reports that Fengzhushou signed an AI computing order in the ¥10 billion range while holding about ¥300 million in cash, with plans to buy more than ¥7 billion of AI servers. The board has also approved guarantees for a subsidiary's borrowings.
• Source: TMTPost · 2026-10-21
• Editor's Take: Order books and balance sheets have never been further apart in Chinese AI. Whoever finances this carries the credit risk, not the customer.

---

III. Humanoid Robots

1. Unitree's H2-X Is a Bodyguard With Eight Hours of Battery

• Summary: Unitree unveiled the H2-X, a humanoid it markets as a personal bodyguard with a maximum runtime of about 8 hours. In the demo it lifts a car and kicks a car door in.
• Source: ITHome · 2026-10-21
• Editor's Take: Lifting a car is a stunt. Eight hours is a shift. Buyers will ask about the second number before the first.

2. Chery's Moja Shows Mobao, a Home Companion at 0.9 Metres

• Summary: Moja Robotics / 墨甲机器人, the robot unit under Chery, launched Mobao, its first consumer home companion humanoid, at 0.9 metres tall with 22 degrees of freedom for walking and body expression. Moja says it has sold more than 3,000 units across 100-plus use cases in 60-plus countries.
• Source: ITHome · 2026-10-21
• Editor's Take: A carmaker selling a companion robot is a channel decision. Chery has service centres and dealers, and that is what a home robot actually needs.

3. China Telecom Spins Out Gamma Robotics for Self-Evolving Humanoids

• Summary: Gamma Robotics / 伽马智能, the first market-facing company incubated by China Telecom's AI research institute TeleAI, registered in Shanghai's Xuhui district. It will work on full-model self-evolving humanoid robots with TeleAI.
• Source: ITHome · 2026-10-21
• Editor's Take: A telecom carrier incubating a robot company follows the same logic as a carmaker doing it. Distribution and deployment sites already exist.

4. Geek+ Says Natural-Language Correction Took OOD Success From 5% to 95%

• Summary: Geek+ / 极智嘉 detailed Gravity Control, a method that lets operators correct a robot in plain language, which it says lifted out-of-distribution task success from 5% to 95%.
• Source: AI Era · 2026-10-21
• Editor's Take: Human correction in natural language is the cheapest supervision there is. The claim is only as good as the size of the OOD test set behind it.

---

IV. Autonomous Driving

1. Nio Says Its World-Model ADAS Is on More Than 950,000 Cars

• Summary: At the World Intelligent Connected Vehicle Conference on Wednesday, Nio CEO Li Bin said the company's driver-assistance stack, built on a world-model architecture with closed-loop reinforcement learning, is deployed on more than 950,000 vehicles, with more than 9.14 billion kilometres of assisted driving recorded.
• Source: ITHome · 2026-10-21
• Editor's Take: Fleet size is the input to a data flywheel, not evidence of autonomy. Nio is publishing the input, which is the honest number to publish.

2. Hesai Drops the Line-Count Race With a 6D Full-Colour Lidar

• Summary: Hesai published details of a next-generation 6D full-colour lidar, claiming native pixel-level alignment of RGB and depth on the chip. Its framing is that autonomy and embodied AI now want high-confidence ground truth more than more scan lines.
• Source: ITHome · 2026-10-21
• Editor's Take: Selling colour instead of line count means selling to data pipelines. That is a sensor market talking to robot training, not to car spec sheets.

3. Momenta's Cao Xudong Puts an L4 Scale Inflection in 2028

• Summary: At the Physical AI and World Models conference this week, Momenta CEO Cao Xudong argued world models are accelerating the autonomy version of Moore's law and predicted 2028 as the inflection point for scaled L4 deployment.
• Source: ITHome · 2026-10-21
• Editor's Take: A 2028 date from a company selling into that market is a planning assumption, not a forecast. The useful part is which inputs it depends on.

4. IDC's First Auto AI Cockpit Report Puts Volcano Engine First

• Summary: IDC tracked AI cockpit solutions as a standalone add-on market for the first time in its China automotive cloud report. Volcano Engine took the largest share in the first half of 2026, with Alibaba second.
• Source: ITHome · 2026-10-21
• Editor's Take: A cockpit assistant is a cloud contract with a steering wheel attached. The first-mover list here is a cloud ranking wearing a car badge.

---

V. World Models / Physical AI

1. Shengshu Opens Its Physical AI Training Run to a Two-Week Livestream

• Summary: Shengshu Technology's world-model team started a public, continuous livestream of the full training run for Motubrain 2.0 on Wednesday, scheduled to last half a month. It follows Xiaomi's MiMo team publishing reinforcement-learning progress in the open.
• Source: QbitAI · 2026-10-21
• Editor's Take: Publishing the training run is marketing that cannot be faked. It also sets a public baseline for how quickly the next model has to improve.

2. A Tsinghua Spin-off Put World Models Into Robots a Year Before LeCun's Take

• Summary: QbitAI profiles Qianjue / 千诀, a Tsinghua-linked company working on latent-variable identifiability in world models, which it says predates comparable LeCun work by a year. The approach leans on physical rules and cross-embodiment reuse rather than predicting video frames.
• Source: QbitAI · 2026-10-21
• Editor's Take: Skipping frame prediction to model physical structure is a narrower bet. It is also the only version that transfers across bodies.

3. Where China Wants to Train Embodied Models: Real Workstation Floors

• Summary: Leiphone surveys the case for "live training platforms" that put robot learning on real workstations. It cites a CAICT report estimating that general embodied models need tens of millions of hours of real interaction data, while globally available high-quality real-robot data sits in the millions.
• Source: Leiphone · 2026-10-21
• Editor's Take: The data gap is the whole argument for training in production. Whoever owns the workstation owns the training set.

---

物界前沿 | EastVoice · 27 items: 8 foundation models, 8 software, 4 humanoid, 4 autonomous driving, 3 world models / physical AI.