Headlines · Hot List · Forum · 中文

EastVoice

PHYSIX FRONTIER EASTVOICE

2026.10.31 · Saturday

Issue 071

FIVE MINUTES A DAY, THE GLOBAL AI SHIFT

¥105.5 billion (~$15.8 billion)

DeepSeek second round from about 30 investors

¥615.5 billion

DeepSeek valuation after second round

20,000

Dexterous hands China shipped last year, per Boya

3,600 pages

Baidu ERNIE Unlimited OCR 2 single-pass demo

Top Stories 3 · dual commentary | Section Picks 24 · one-line takes

TOP STORIES

DeepSeek's ¥105.5 billion round, AgiBot's GO-3 Preview benchmark claims, and Baidu's open-sourced OCR model reading 3,600 pages in one pass.

1AgiBot's GO-3 Preview claims three benchmark crowns, and a shift at the convenience store

AgiBot released GO-3 Preview on October 29, built on an embodied-native WRAM (World Reasoning Action Model) architecture that ties task reasoning, future prediction and action generation into one loop. The company says it tops three public leaderboards. Demo footage shows it making jianbing and working a convenience-store counter.

What it means for you|Demo footage shows the model making jianbing and working a convenience-store counter, but the take notes a benchmark result is not yet a product that runs 400 times.

Editor's take|A model that can fry an egg on camera is a benchmark result. A model that can do it 400 times before the griddle needs cleaning is a product.

Source: ITHome (2026-10-30)

2SynDexy ships its first batch of dexterous humanoids to factory floors

SynDexy delivered the first small batch of its SynDexy C1 humanoid, moving from lab validation to engineering delivery. The units go into precision assembly lines at consumer-electronics and component plants.

What it means for you|The units go into precision assembly lines at consumer-electronics and component plants, moving from lab validation to engineering delivery; the next data point is customer reorders.

Editor's take|Small-batch delivery is where purchase orders replace press releases. The next data point is how many units the customer reorders.

Source: Leiphone (2026-10-30)

3Tsinghua, Infinigence and AgiBot teach robots to learn from each other

The three organisations proposed Collective Experience Learning, a training paradigm where robots share experience instead of learning only from human demonstrations. They validated it on a cluster of 96 real robots.

What it means for you|Robots share experience instead of learning only from human demonstrations, validated on a cluster of 96 real robots, which the take calls the cheapest scaling law in robotics.

Editor's take|Ninety-six robots sharing one lesson is the cheapest scaling law in robotics, provided the lesson transfers across different hardware.

Source: QbitAI (2026-10-30)

SECTION PICKS

24 AI stories worth your time, each with a one-line take.

Humanoid RobotsAt IROS 2026, dexterous hands cleared the demo bar and hit the reliability wall

This year's IROS floor was full of hands threading beads, inserting USB plugs, spinning pens and mixing cocktails. Leiphone's takeaway from ICRA, WRC and IROS this year is that complex motion is no longer the hard part.

What it means for you|The IROS floor showed hands threading beads, inserting USB plugs, spinning pens and mixing cocktails, but the take says production asks for eight hours, five days a week, with a maintenance log.

One-line take|Every one of those demos ran for ninety seconds. Production asks for eight hours, five days a week, with a maintenance log.

Source: Leiphone (2026-10-30)

Humanoid RobotsBoya's founder: 20,000 dexterous hands sold last year, most now gathering dust

Liu Xin of Boya Intelligence estimates China shipped about 20,000 dexterous hands last year. Most went into humanoid stage shows and were then shelved. Cheap low-DOF units perform, high-DOF units break, and the expensive ones stay in the box. His view is that after GPT-6, hardware is what holds robots back.

What it means for you|Most of the roughly 20,000 dexterous hands shipped last year went into humanoid stage shows and were then shelved, which the take calls a demand problem, not a dexterity problem.

One-line take|An industry that sells hands for performances has a demand problem, not a dexterity problem.

Source: Leiphone (2026-10-30)

Humanoid RobotsThe embodied funding squeeze is showing up in the secondary market

TMTPost reports that AgiBot's Hong Kong listing has slowed, some holders are offering old shares at a 50% discount with no buyers, and several companies have stopped approving share reforms to keep stock from leaking out. Investors describe the pressure as concentrated after November.

What it means for you|Some holders are offering old shares at a 50% discount with no buyers, and the take says the market is asking which companies have revenue that isn't a demo contract.

One-line take|A 50% haircut on paper is the market's way of asking which of these companies has revenue that isn't a demo contract.

Source: TMTPost (2026-10-30)

Humanoid RobotsA deployment league table for robots that already work in factories

A survey of robots that have entered real production lines looks at cycle time, failure rates and payback periods rather than demo footage. Its finding is that one successful grasp says little about keeping pace with a line, and one installed unit says little about whether the package replicates.

What it means for you|The survey looks at cycle time, failure rates and payback periods rather than demo footage, finding one installed unit says little about whether the package replicates.

One-line take|The industry has plenty of videos and very few maintenance logs. The logs are where the market will be decided.

Source: Huxiu (2026-10-30)

Autonomous DrivingStelato's V8 puts Huawei's L3 architecture in a ¥329,800 sedan

Huawei's HarmonyOS Smart Mobility lineup published options for the Stelato V8. It ships with ADS 5 as standard, ±12° rear-wheel steering across the range, and an Ultra trim built on an L3 autonomous architecture. Pre-sale starts at ¥329,800, with launch expected in early November.

What it means for you|The Stelato V8 ships with ADS 5 as standard and an Ultra trim built on an L3 autonomous architecture, pre-sale from ¥329,800, with launch expected in early November.

One-line take|L3 hardware in a ¥330,000 car is easy. The liability contract behind it is the part nobody has published.

Source: ITHome (2026-10-30)

Autonomous DrivingHorizon's HSD stack is heading into BYD's God's Eye B

BYD's November models will carry a new God's Eye B configuration combining Horizon Robotics' Journey 6P chip with the HSD software stack, according to Chinese media cited by ITHome. Denza, Han and high-end Ocean models are expected to follow.

What it means for you|BYD's November models will carry God's Eye B combining Horizon's Journey 6P chip with the HSD stack, which the take calls a domestic chip vendor's largest volume reference.

One-line take|BYD just handed a domestic chip vendor its largest volume reference. That is worth more than any benchmark score.

Source: ITHome (2026-10-30)

Autonomous DrivingZhuoyu takes its driving stack to trucks, and promises a one-year payback

Zhuoyu Technology launched Panxing, a commercial-vehicle assistance system split into PDS 4 for highway NOA on heavy trucks and PDS 5 for buses, light trucks and robovans. The company claims operators can recover the cost within a year.

What it means for you|Zhuoyu's Panxing splits into PDS 4 for highway NOA on heavy trucks and PDS 5 for buses, light trucks and robovans, with a claimed one-year payback operators can recover.

One-line take|Trucking fleets buy on fuel and insurance math, not on autonomy narratives. A one-year payback claim will be audited by CFOs, not analysts.

Source: Leiphone (2026-10-30)

World Models & Physical AIIM's LS8 puts a world model on a ¥250,000 SUV

SAIC's IM launched the LS8 from ¥250,000, with a steer-by-wire chassis and CATL batteries standard across the range. Momenta's world model runs on the car, part of a pitch that includes lifetime AI features.

What it means for you|The LS8 launches from ¥250,000 with a steer-by-wire chassis and CATL batteries standard, and Momenta's world model runs on the car as part of lifetime AI features.

One-line take|World models are now a trim-level feature. The interesting question is what happens to the car's behaviour when the model is updated six months after purchase.

Source: QbitAI (2026-10-30)

World Models & Physical AILi Fei-Fei's three pits, and the $8.2 billion anchor under the world-model trade

A Leiphone profile traces Fei-Fei Li's path from ImageNet to spatial intelligence and the three gaps she has to close. It notes that $8.2 billion in capital has already set a valuation anchor for the world-model sector.

What it means for you|The profile notes $8.2 billion in capital has already set a valuation anchor for the world-model sector, while the take says spatial intelligence lacks ImageNet's cheap labels and narrow task.

One-line take|ImageNet worked because the labels were cheap and the task was narrow. Spatial intelligence has neither advantage.

Source: Leiphone (2026-10-30)

World Models & Physical AIA Hangzhou data centre wants to compress two years of embodied data into one month

Yingsu opened what it calls China's first embodied simulation data-collection centre, in Hangzhou. The industry needed roughly two years to accumulate a million hours of robot data. Yingsu says it can produce that in a month.

What it means for you|Yingsu says it can produce a million hours of robot data in a month, where the industry needed roughly two years, though the take warns of greasy kitchen counters.

One-line take|Synthetic hours are cheap. The bill arrives when a policy trained in simulation meets a greasy kitchen counter.

Source: Leiphone (2026-10-30)

Foundation ModelsDeepSeek raises ¥105.5 billion at a ¥615.5 billion valuation

DeepSeek's second round brought in ¥105.5 billion (~$15.8 billion) from about 30 investors, according to reports cited by TMTPost. State capital accounts for 27%, VCs 26.5% and tech companies 41.9%, including Tencent, CATL, CXMT, SMIC, Meituan and vivo. It is the largest venture round ever raised by a Chinese AI company.

What it means for you|The round drew about 30 investors including Tencent, CATL, CXMT, SMIC, Meituan and vivo, with state capital at 27%, the largest venture round ever raised by a Chinese AI company.

One-line take|A cap table that mixes chip fabs, battery makers and a video platform is an industrial policy instrument, not a venture round.

Source: TMTPost (2026-10-30)

Foundation ModelsSiliconFlow files for a Hong Kong listing

Token supplier SiliconFlow has had its Hong Kong IPO filing accepted by the CSRC. The company reports over 10 million registered users and 13,000 enterprise customers, and raised close to ¥2.9 billion this year. Frost & Sullivan ranks it as China's largest independent token supplier.

What it means for you|SiliconFlow's Hong Kong IPO filing was accepted by the CSRC, reporting over 10 million registered users and 13,000 enterprise customers, and the take says the filing will show its margins.

One-line take|Routing tokens is a thin-margin business until you own the capacity underneath. The filing will show which side of that line SiliconFlow sits on.

Source: ITHome (2026-10-30)

Foundation ModelsDeepSeek's front-end code leaks a cybersecurity model called Cyber

Developers found application routes and a feature gate for a Cyber model in DeepSeek's site code, pointing to enterprise and personal editions in limited grey release. DeepSeek has not announced the product.

What it means for you|Developers found application routes and a feature gate for a Cyber model in DeepSeek's site code, pointing to enterprise and personal editions in limited grey release, though DeepSeek has not announced it.

One-line take|A security model is the one product where a leak before launch is itself a security story.

Source: ITHome (2026-10-30)

Foundation ModelsBaidu open-sources an OCR model that reads 3,600 pages in one pass

Baidu released ERNIE Unlimited OCR 2 on October 27, removing the page ceiling that constrains document parsers. In its demo it processed all seven Harry Potter books, more than 3,600 pages, in a single pass.

What it means for you|Baidu's ERNIE Unlimited OCR 2 removes the page ceiling that constrains document parsers, processing all seven Harry Potter books, more than 3,600 pages, in a single pass.

One-line take|Long-context OCR is unglamorous and immediately useful. That combination is rare in Chinese model releases this year.

Source: ITHome (2026-10-30)

Foundation ModelsXiaomi open-sources a 0.8B OCR model for documents and visual Q&A

Xiaomi released a 0.8B parameter OCR model that handles document parsing and visual question answering in one model, with Markdown and JSON output.

What it means for you|Xiaomi's 0.8B parameter model handles document parsing and visual question answering with Markdown and JSON output, which the take says helps phone makers avoid per-token cloud rent.

One-line take|Small models on-device are how phone makers avoid paying per-token rent to a cloud they do not control.

Source: ITHome (2026-10-30)

Foundation ModelsIluvatar's Tiangai 300 matches H200 in hands-on tests

At a media day this week, Iluvatar CoreX benchmarked its Tiangai 300 against Nvidia's H200 and reported comparable performance. The company is now pitching heterogeneous systems and an agent-based software stack rather than raw FLOPS.

What it means for you|Iluvatar benchmarked its Tiangai 300 against Nvidia's H200 and reported comparable performance, now pitching heterogeneous systems and an agent-based software stack rather than raw FLOPS.

One-line take|Matching H200 on one workload is a milestone. Matching CUDA's decade of tooling is the actual product.

Source: Leiphone (2026-10-30)

Foundation ModelsQwenputer puts a 32-bit RISC-V CPU inside a language model

Developer Divyansh Agrawal released Qwenputer on October 22, built on Qwen3.5-4B. The model emulates a 32-bit RISC-V CPU and memory inside its own network, so it can execute programs without calling Python or any external tool.

What it means for you|Qwenputer emulates a 32-bit RISC-V CPU and memory inside its own network, so it can execute programs without calling Python or any external tool.

One-line take|Teaching a model to be its own interpreter is a neat trick. Whether it computes arithmetic more reliably than a calculator is a separate question.

Source: ITHome (2026-10-30)

AI SoftwareApple Intelligence is finally clearing Chinese regulatory ground

Apple said on October 31 that the first batch of Apple Intelligence features aimed at improving the iPhone experience is coming to users in China, without giving a full feature list or a date for every capability.

What it means for you|Apple said the first batch of Apple Intelligence features aimed at improving the iPhone experience is coming to users in China, without a full feature list or dates for every capability.

One-line take|Two years after launch, Apple's China AI story is still a schedule announcement. Domestic rivals shipped on-device assistants while it waited.

Source: ITHome (2026-10-31)

AI SoftwareHuawei's XiaoYi gets memory and triggers, with 100+ apps wired in

Huawei upgraded XiaoYi on the Mate 90 series, with two new capabilities: XiaoYi Bangji, which remembers context across apps, and trigger-based AI that fires on events. More than 100 apps have integrated the APIs.

What it means for you|Huawei's XiaoYi Bangji remembers context across apps and trigger-based AI fires on events, with more than 100 apps having integrated the APIs, so the phone acts without being asked.

One-line take|Trigger-based agents are the pragmatic version of the assistant dream. The phone acts without being asked, which is also the version users will complain about.

Source: ITHome (2026-10-30)

AI SoftwareBaidu rebuilds search around recursive self-improvement, and cuts API prices

Baidu announced four AI upgrades to its app on October 30, including a search stack it describes as recursively self-evolving, a rebuilt search API priced at ¥4.9 per 1,000 calls during the campaign, and new camera and GUI features. That undercuts DeepSeek's ¥5 standard rate.

What it means for you|Baidu's rebuilt search API is priced at ¥4.9 per 1,000 calls during the campaign, undercutting DeepSeek's ¥5 standard rate, alongside new camera and GUI features.

One-line take|Baidu is pricing search calls below a competitor that does not sell search. That tells you where the margin is going.

Source: QbitAI (2026-10-30)

AI SoftwareXiaohongshu buys travel-planning app Pi Travel outright

Registration records show Xiaohongshu has become the sole shareholder of the company behind Pi Travel, an AI itinerary app, with founder Zheng Yifan exiting. Registered capital is ¥1 million.

What it means for you|Registration records show Xiaohongshu became sole shareholder of the company behind Pi Travel, an AI itinerary app, with founder Zheng Yifan exiting and registered capital of ¥1 million.

One-line take|A travel planner is the shortest path from a saved post to a booked trip. Xiaohongshu bought the last mile.

Source: TMTPost (2026-10-30)

AI SoftwareShagang puts Qwen agents into electric-furnace and finance workflows

Jiangsu Shagang Steel has deployed Qwen Office agents in production, covering electric-furnace review, transformer condition assessment and financial analysis. The company says some core scenarios are now hundreds of times faster.

What it means for you|Shagang deployed Qwen Office agents covering electric-furnace review, transformer condition assessment and financial analysis, and says some core scenarios are now hundreds of times faster.

One-line take|Steel mills are a better AI market than law firms. The workflows are documented, the data is instrumented, and nobody argues about creative license.

Source: Leiphone (2026-10-30)

AI SoftwareAlibaba Cloud's Qoder passes $100 million in recurring revenue

Qoder, Alibaba Cloud's AI coding product, has crossed $100 million in annualised recurring revenue, and the team is splitting the product line. The number was first discussed at the September Yunqi conference.

What it means for you|Qoder crossed $100 million in annualised recurring revenue and the team is splitting the product line, making coding assistants the one Chinese AI category with paying enterprise customers at scale.

One-line take|Coding assistants are the one Chinese AI category with paying enterprise customers at scale. Qoder is now the number to beat.

Source: TMTPost (2026-10-30)

AI SoftwareTwo Chinese AI companion apps vanish from the US App Store

In mid-October, PolyBuzz, owned by Zuoyebang, was pulled from the US and UK App Stores on October 15, and a second Chinese AI companion app disappeared from the US store at the same time. The products were reportedly earning over ¥100 million a month.

What it means for you|PolyBuzz was pulled from the US and UK App Stores on October 15 and a second Chinese AI companion app disappeared from the US store, after reportedly earning over ¥100 million a month.

One-line take|Companion apps were the fastest Chinese AI export to make real money abroad. Apple just showed how quickly that channel can be closed.

Source: Huxiu (2026-10-30)

📖 The full EastVoice issue on the Chinese forum

EastVoice — five minutes a day, the global AI shift
Physix Frontier (Shenzhen) · covering foundation models, AI software, humanoid robots, autonomous driving and physical AI