Legal name: Qianjue Technology
Qianjue Technology Deep Dive: How a Body-Agnostic Embodied Brain Pairing the Predictive World Model Polibrain with Brain-Inspired Partitioning and Polynomial Representation Reached a Hundred Thousand Cleaning and Service Robots
Qianjue builds a body-agnostic embodied brain from the predictive world model Polibrain and brain-inspired partitioning; nine funding rounds ending in an A+ round in September 2026, a Yangzhou Data Industrial Park base signed in October, and a customer base of which media report nearly half sits in Jiangsu.
Executive Summary
Qianjue Technology (Beijing Qianjue Technology Co., Ltd.) turns a predictive world model into a robot decision layer: it builds no bodies, only an "embodied brain" reusable across form factors. Founded on June 13, 2023 in Beijing's Chaoyang district and incubated at Tsinghua University's Center for Brain-Inspired Computing Research, it is led by founder and CEO Gao Haichuan, a Tsinghua automation PhD who from 2018 headed Professor Chen Feng's brain-inspired dual-arm robotics group. Chief technical advisor Chen Feng is among China's first brain-inspired computing researchers and has led multiple major projects under the China Brain Science Program. Co-founder and general manager Xiao Haimo came from Midea's smart-home business group. The core technology is Polibrain, a brain-inspired partitioned predictive world model with a loosely coupled causal representation, formally released on October 20, 2026 at the inaugural Physical AI and World Model Conference in Nanjing. Commercially the company reports cumulative device connections at the hundred-thousand scale, concentrated in semi-service, semi-household cleaning — it supplies task-scheduling and scene-understanding models to Dreame's robot vacuum line — and extending to hotel cleaning, restaurant service, indoor precision operation and smart delivery. Media reporting in October 2026 puts close to half of the customers in Jiangsu, naming Tiance, Xingong Group and Yika in Nanjing and Dreame, Magic Atom and Pangolin in Suzhou. On 20 October 2026 in Nanjing the company signed with the Yangzhou Ecological Technology Innovation City Administrative Committee, settling a world-model and robot-brain Yangtze River Delta base in the Yangzhou Data Industrial Park; the media puts the Yangzhou headcount above forty while enterprise registry data show zero insured employees for that entity and 37 for the Beijing parent in 2025, and the one-billion-item data lines and fifty-plus deployment tasks described for Yangzhou are stated in the future tense, so they read as plans rather than capacity in operation. Nine funding rounds have closed since 2024 by the media count, the latest being an A+ round of several hundred million RMB on 8 September 2026 backed by Yuanhe Houwang, Xiaoguang Capital, Xinneng Venture, Inno Angel Fund, Jingming Capital, Weilai Bianji and Maple Pledge Fengcheng Capital; no standalone A round appears in the sequence, the company disclosed no post-money valuation, and local media relay an unconfirmed figure of about CNY 7 billion for the Beijing headquarters.
Caveats: the company is private with no audited financials; the hundred-thousand figure, customer list and dataset scale are company claims, and the hundred-thousand device number appears both as achieved and as a 2026 target at different points, adjudicated in Discrepancies; the predictive world model has no public benchmark; the roughly CNY 7 billion valuation relayed by local media is a reporter's account without company confirmation, while the A+ announcement discloses no post-money valuation, and the Yangzhou headcount, data-line and deployment-task figures are media and planning statements rather than audited or operating data.
The verdict: the sharpest technical thesis and the most counter-intuitive commercialization in the embodied-AI field. It is sharp because the company publicly rejects two industry conventions — it does not chase generative world models (no next-frame prediction, only state evolution) and does not chase humanoid bodies (one brain, many bodies, form follows scenario). It is counter-intuitive because its first orders came from robot vacuums and hotel cleaning rather than stage demos. The founder's much-quoted judgment: "There is no ChatGPT moment in the robotics industry" — no single technical breakthrough will detonate the whole sector at once. That claim is both the reason for its incremental path and the premise any evaluation must accept.
Company Overview
Founded June 13, 2023 in Chaoyang, Beijing, the company was incubated at Tsinghua's Center for Brain-Inspired Computing Research with support from China's Brain Science Program. The team is predominantly Tsinghua automation PhDs and master's graduates with backgrounds in brain-inspired computing, robot control and machine learning; members have published in NeurIPS Spotlight and TPAMI and placed third in the ViZDoom international reinforcement-learning competition.
The founding team pairs academia with industry. Gao Haichuan holds both a bachelor's and a PhD from Tsinghua's automation department and from 2018 led the brain-inspired dual-arm robotics group in Chen Feng's lab, taking multiple autonomous dual-arm robots from zero to one over more than six years. CTO Zhang Tianren owns the world-model theory and algorithms; his paper When Do Neural Networks Learn World Models? won the outstanding paper award at the ICLR 2025 world-model workshop and was the first to establish a rigorous mathematical definition for world-model learning. Co-founder and general manager Xiao Haimo was a senior director of strategy management in Midea's smart-home group with full-chain R&D, manufacturing, sales and service experience at a Fortune 500 firm, responsible for commercialization and industry resources.
The team's technical starting point predates the company. Multiple reports note that by 2017 and 2018, while still in the Tsinghua lab, the group had already committed to the predictive world model line and placed well in international competitions; the company later did not foreground this history until its route was discussed alongside LeCun's in 2026. The timeline matters: the bet on prediction is not a 2026 bandwagon but a path dependency of more than seven years.
The company positions itself in the software and model layer of the value chain: no bodies, but deployable, mass-producible and continuously evolving intelligence sold to robot makers of different categories, sparing them duplicate R&D. That position means commercialization need not wait for humanoid volume — customers already ship hardware, so algorithms can reach real devices earlier.
Yangzhou Base and the Yangtze River Delta Footprint
At the inaugural 2026 Physical AI and World Model Conference in Nanjing on 20 October 2026, Qianjue Technology signed on site with the Yangzhou Ecological Technology Innovation City Administrative Committee, and the "Qianjue Technology World Model and Robot Brain Yangtze River Delta Base" was formally settled in the Yangzhou Data Industrial Park. Press reports do not give the full registered name of the signing party; enterprise registry data identify the entity that has landed as Qianjue Technology (Yangzhou) Co., Ltd. (91321016MAKJL2RW4A, incorporated on 10 July 2026, registered capital CNY 30 million, 21/F Tower G, Yangzhou Software Park, legal representative Gao Haichuan, wholly owned by Beijing Qianjue Technology).
Everything described on the Yangzhou side is stated as a plan. The local media wording is entirely in the future tense: the data segment "will build a production line handling one billion data items a year", the company "wants to form a collection and annotation line producing one billion data items a year", and it "will act as a delivery centre, taking on deployment and commissioning tasks for more than fifty robot models". These are targets rather than capacity in operation, and the report records them as such. A second "more than fifty" carries a different meaning: coverage that the embodied brain "fully adapts to more than fifty robot models across seven major categories including bipedal humanoids and wheeled platforms" counts product-adapted machine types, which is a different quantity from the delivery-centre deployment tasks.
On headcount, the media figure is more than forty employees in Yangzhou, mostly recruited locally with a management team brought from the headquarters, while enterprise registry data show an insured headcount of zero for Qianjue Technology (Yangzhou) and 37 for Beijing Qianjue Technology in 2025. The two bases are recorded side by side and are not converted into each other.
Local media also report that the headquarters has moved south to Nanjing's Xuanwu district, at the Xuzhuang High-Tech Zone and the Jiangsu International Data Port; that headquarters and the Yangzhou base coexist and are not the same entity.
The valuation needs its own label. Local media relay that "according to available information, Beijing Qianjue Technology, which works on embodied-intelligence brain systems, is currently valued at about CNY 7 billion". The wording is a reporter's account, with no confirmation from the company or from Gao Haichuan, and the subject is the Beijing headquarters rather than the Yangzhou subsidiary, while the A+ round announcement states that the specific use of proceeds and the post-money valuation are undisclosed. The report therefore records a media-reported valuation of about CNY 7 billion for the Beijing headquarters, unconfirmed by the company, with the A+ announcement disclosing no valuation.
Technology
Route Choice: Predictive versus Generative World Models
World models split into two routes. The generative route predicts change by generating future frames (video generation and 3D scene generation belong here); the predictive route forecasts the evolution of physical state and outputs low-dimensional abstract features without reconstructing pixels. Qianjue chose the latter on compute economics: rebuilding every pixel spends capacity on lighting and material irrelevant to the task while neglecting the governing physics. Gao's analogy is ball sports — a player swinging does not render crisp frames in the head but relies on a low-dimensional prediction of the ball's trajectory.
The engineering consequence is that the model must learn state transitions inside a task-relevant latent space, which is precisely the hard part: pick the wrong representation and even accurate prediction will not transfer. Qianjue's answer is to solve representation separately (polynomial representation) and to separate prediction from execution (the loosely coupled architecture). Polibrain is therefore not a model that generates worlds but a map that enables planning — Gao's illustration is a merchant shipping lychees from Guangzhou to Chang'an: one solution is to walk every possible route (the current VLA trial-and-error route), the other is to draw a map first and then plan.
The Brain-Inspired Partitioned Architecture
Brain-inspired partitioning is a third path alongside the mainstream layered and end-to-end approaches: it mimics the functional division of the human brain, splitting complex environmental information into relatively independent modules (vision, hearing, decision, interaction, memory), modeling and predicting each separately before high-level fusion. Officially described as a decomposed, decoupled large-organization structure with linear pretraining, the company calls it the first brain-inspired design of a multimodal large model (public statements differ between ICLR 2023 and ICLR 2024; see Discrepancies).
The architecture yields three benefits: parallel multi-module processing raises sample efficiency, cuts training cost and speeds inference; decoupling prevents the end-to-end architecture from being bound to a body form; and the stack from silicon to algorithm is fully self-controlled. The third deserves note — the on-device "Brain Dock" supports both spiking neural networks and conventional artificial networks, runs on NVIDIA and AMD parts as well as Huawei Ascend and Lynxi brain-inspired chips, and claims materially lower power than GPU architectures for privacy-sensitive or connectivity-poor settings.
Polynomial Representation and Effective Degree
Polynomial representation is Qianjue's deepest and least externally verifiable claim. The problem it targets is feature contamination and spurious causality: models latch onto background texture, lighting or surrounding clutter as evidence, and more insidiously mistake correlation for causation — the owner leaves, takes the keys, and the keys vanish from the drawer, so the model learns that the owner leaving makes the keys disappear; when the owner actually forgets the keys and asks the robot to find them, the robot concludes they cannot be in the drawer.
Qianjue's answer is a mathematical method for extracting the variables that truly drive task outcomes. On theory there are two pieces: a 2025 ICML paper with Tsinghua, When Do Neural Networks Learn World Models?, which starts from multitask learning and low-order preference and uses Fourier-Walsh analysis of function structure over discrete Boolean worlds to explain when a model can identify true latent variables; and a 2026 follow-up introducing Effective Degree, which walks interpolation paths between real data points, fits outputs with orthogonal polynomials and weights components by order — low-order dominance means a small ED, more and larger high-order components mean a larger ED. The value is that it is computable, comparable and can be pushed into training to penalize needless complexity.
One boundary must be stated plainly: LeCun's group published When Does LeJEPA Learn a World Model? in May 2026, also examining latent identifiability and low-order preference. Qianjue asserts its work came about a year earlier and that the two criteria are partly equivalent. But earlier is not better — the mathematical tools (causal inference over Boolean spaces versus a temporal framing) and the objects of study differ, and superiority needs independent assessment; this report records only the chronology.
Causal Memory and the Continuous Decision Loop
In continuous tasks objects get occluded and some state cannot be judged from the current frame. Qianjue adds counterfactual-inference-based history filtering: it decides whether a historical observation will affect later prediction and decision, then keeps or discards it. The robot need not remember every frame — background colour does not matter, which cabinet the cup went into does.
On top sits a continuous decision loop: each control command is accompanied by an internally generated expected feedback; when execution deviates, the model traces the source of deviation within milliseconds (bad grasp, slip, unexpected obstacle), re-evaluates the global goal and adjusts. The company's phrasing is predict, verify, admit error, self-heal. In demonstrations this appears as task resumption — after an interruption such as yielding to a pedestrian or taking a temporary instruction, the robot remembers the interruption point and resumes the original task.
Loosely Coupled Architecture and One Brain, Many Bodies
Polibrain separates the shared representation of environment and task from the execution capability of a specific robot. The upper general intelligence base handles task understanding, environment representation, physical state prediction and causal reasoning, independent of body shape; the lower hardware adaptation interface connects to different control systems, motion primitives and trajectory planning. Gao's example is concrete: to load laundry, a humanoid can hold the clothes with one hand and open the door with the other, while a single-arm wheeled robot can ask the washing machine to open itself through smart-home IoT — different hardware and execution, the same change in the physical world.
The company says it has adapted nearly 50 robot configurations (an earlier figure was twenty-plus), spanning industrial arms, wheeled bases, dexterous multi-finger hands, bipedal humanoids, quadrupeds, drones and robot vacuums.
Technical Shortfalls
First, the predictive route lacks a public benchmark: how accurate state prediction is and how reliable the loop is in complex scenes rests on company videos and qualitative description. Second, polynomial representation and effective degree are theoretical tools with no controlled experiment showing measurable generalization gains on real tasks. Third, the robustness of the pure-vision approach (no lidar) under glare, low light and specular reflection is undisclosed. Fourth, being a year ahead of LeCun's line is chronology, not a performance result. Fifth, data on actual deployment share of brain-inspired chips, compute specifications and comparisons against GPU schemes is missing.
Product and Commercial Model
Four Product Lines
The embodied brain (perception-decision-control model) is the core: it plans tasks without language instructions or prompts, its decoupled architecture spans form factors, and pure vision plus causal reasoning enables implicit spatial modeling and task planning in home and service settings.
Polibrain OS is the cloud development platform: API and web access, a perception model suite covering object recognition and environment understanding, and communication with body control systems through motion primitives. The company says the unified perception layer is engineered and entering external validation, exploring one perception stack across many robot forms.
The Brain Dock is the on-device compute platform: spiking and artificial networks, NVIDIA and AMD as well as domestic brain-inspired silicon, pitched on low power and local compute for privacy-sensitive or poorly connected sites.
The data-model integrated service covers collection, cleaning, labelling, training and fine-tuning for customer scenarios, with in-house annotation craft for robot-specific data such as task-decomposition sequences and grasp trajectories; the data team has served autonomous-driving 3D point-cloud labelling, delivery keypoint labelling and household object collection.
Commercialization Path: Starting from Vacuums and Hotel Cleaning
This is the choice most worth recording. The first orders were not humanoids but food-service, cleaning and hotel-service robots with established demand. Gao's formulation: customers do not care about your architecture or algorithm, they care whether the thing can tidy the slippers strewn around a hotel room.
The largest deployment category is semi-service, semi-household cleaning: Qianjue supplies task-scheduling and scene-understanding vision models to Dreame's robot vacuum business, and has deployed in hotel cleaning, restaurant service, humanoid reception and indoor precision operation, with some projects at scale. Midea is also named as an industry partner. Media coverage in October 2026 says close to half of the customers come from Jiangsu, naming Tiance, Xingong Group and Yika in Nanjing and Dreame, Magic Atom and Pangolin in Suzhou; those names extend the customer list beyond the two firms that were previously disclosed. The company reports cumulative device connections at the hundred-thousand scale and said in late 2025 that it expected to exceed one hundred thousand devices within 2026 — two different time points that must be read separately.
The vacuum supply chain matters more than it appears: robot vacuums are one of the few categories shipping at volume, so entering it means the technology is stress-tested and iterated against dense real-world data — density no stage demo can supply. Conversely, the task complexity of vacuuming is far below tidying a room, so using it to prove a general embodied brain spans a wide gap, and that is the first question any evaluation should press.
Deployment Claims
The company lists four deployment advantages: no environment modification (against schemes needing QR codes and elevator retrofits, claiming over 90 percent lower deployment and retrofit cost); proactive intelligence that spots work (after wiping a table, noticing floor litter and starting a cleanup unprompted); task resumption after interruption; and deep understanding of vague instructions such as tidy the room, decomposed autonomously into an execution sequence. Together with pure vision cutting hardware cost, these form the commercial pitch — all company claims without third-party customer quantification.
Competitive Landscape
Rivals fall into three classes. Body makers with in-house brains (Zhiyuan, Unitree) enjoy software-hardware integration and data loops but do not supply brains externally. Independent brain companies (Galbot, Xingdong Jiyuan, Zhongke Diwiji, Westlake Robotics) compete on approach; Qianjue's differentiation is its pure no-bodies position plus the predictive route. World-model research companies (AMI Labs, World Labs, Feijiekosi) target more general world models, while Qianjue narrows the scope to the robot decision layer. Media describe Qianjue as holding the leading market share in its segment; no third-party report supports that figure, so this report marks it unverified and does not use it as evidence.
Against named peers on this list: Zhongke Diwiji uses action optical flow to unify behavior representation across bodies and holds a CVPR ARNOLD challenge title as third-party evidence; Westlake Robotics pairs a WR1 brain with a GAE cerebellum and builds its own hardware; Qianjue attacks the same two problems — cross-body reuse and few-shot adaptation — with polynomial representation and a loosely coupled causal architecture, but has no body business at all and commercializes earlier into lower-complexity categories.
Industry judgments are worth recording: Pudu Robotics founder Zhang Tao said at BEYOND Expo 2026 that embodied AI's ChatGPT moment is at least three to five years away, and Matrix Superintelligence founder Zhang Haixing put the consumer breakout between 2028 and 2030. That aligns with Qianjue's no-ChatGPT-moment thesis: the consensus timeline is shifting from imminent explosion to gradual landing.
Financing and Capital Structure
Six rounds carry names in the media sequence, while the company and media put the cumulative count at nine: an angel round in September 2024 of tens of millions of RMB (Inno Angel Fund, TusStar, the Shuimu Tsinghua Alumni Seed Fund and Lingyi Venture among them); angel-plus in December 2024 of tens of millions, with the same group continuing; Pre-A in March 2025 of tens of millions (Zhuichuang Venture and DT Capital co-leading, with Jingye Intelligent Industry Fund and Zhengjing Zhiyuan investing strategically); Pre-A+ in May 2025 (Junshan Investment, Vertex Ventures, Shixi Capital under GigaDevice); Pre-A++ in December 2025 of nearly RMB 100 million; and the A+ round of several hundred million on 8 September 2026 (Yuanhe Houwang, Xiaoguang Capital, Xinneng Venture, Inno Angel Fund, Jingming Capital, Weilai Bianji and Maple Pledge Fengcheng Capital). No standalone A round appears in this sequence: earlier coverage labelled the same September 2026 financing an A round, which is a naming divergence over a single event (discrepancy 13). Baidu Baike's timeline separately records a Pre-A++ extension in February 2026, so the count of nine cannot be reconciled round by round against the named sequence. Cumulative funding is in the several-hundred-million range; the A+ announcement states that the specific use of proceeds and the post-money valuation are undisclosed, while local media relay an unconfirmed figure of about CNY 7 billion for the Beijing headquarters (discrepancy 9).
The investor mix deserves note: angel-round institutions recur across the sequence, with Inno Angel, TusStar, the Shuimu Tsinghua Alumni Seed Fund and Lingyi Venture all appearing more than once; Pre-A was co-led by Zhuichuang Venture and DT Capital; Pre-A+ added Junshan Investment, Vertex Ventures and Shixi Capital under GigaDevice; the A+ round brought Yuanhe Houwang from the Suzhou Yuanhe system, Xiaoguang Capital, Xinneng Venture, Jingming Capital, Weilai Bianji and Fengcheng Capital. Earlier coverage of the September 2026 financing, then labelled an A round, also listed provincial state capital such as Shandong New Momentum and Shandong Caijin, partly overlapping the A+ disclosure, which is a naming divergence over a single event (discrepancy 13). State capital usually binds to industrial deployment in cleaning, property and public services, which matches the service-first path. Fengcheng Capital has long served as financial advisor and joined the A+ round.
DISCREPANCIES
1. Whether one hundred thousand devices is achieved or targeted. Source A (after the June 2026 A round): cumulative device connections have reached the hundred-thousand scale. Source B (December 2025, Pre-A++): the company expects device count to exceed one hundred thousand within 2026. The time points differ and may reflect a later retrospective of an in-year achievement, but the counting basis (activated units, shipped units, vacuums included?) is undefined. Adjudication: record both time points and flag the basis as unverified. Source grade: OFFICIAL.
2. Round count and named sequence. The company says nine rounds, and six of them carry names in the public record, so the remaining rounds inside the nine are not disclosed one by one. Dates for the early rounds also diverge: Baidu Baike dates the angel round October 2, 2024, Phoenix dates it November 13, 2024, and the current media sequence places it in September 2024; one source puts angel-plus on December 26, 2024 and another says it closed on January 3, 2025. Adjudication: use the named sequence as the skeleton, keep the nine-round media count, and flag the date divergence. Source grade: DB versus MEDIA conflict.
3. The year of the brain-inspired multimodal model. The official site says the world's first brain-inspired multimodal large model at ICLR 2023; a corporate profile says ICLR 2024; other reports cite the CTO's ICLR 2025 world-model workshop outstanding paper. Three different referents — model architecture, conference paper, workshop award — are used interchangeably. Adjudication: record separately, do not merge. Source grade: OFFICIAL (internally inconsistent).
4. The nature of being a year ahead of LeCun. The company and media stress that the 2025 ICML paper precedes LeCun's May 2026 paper by about a year and that the criteria are partly equivalent. Earlier is a chronological fact; equivalence is a self-assessment without third-party review. Adjudication: record the chronology; mark equivalence as a company claim. Source grade: MEDIA and OFFICIAL.
5. Body adaptation counts. Earlier coverage says twenty-plus embodied form factors; the October 2026 Polibrain launch says nearly 50 configurations. Adjudication: present chronologically, with the latter as the latest. Source grade: MEDIA and OFFICIAL.
6. Exclusivity of the largest pure real-world household dataset. The company claims the world's largest purely real-collected household scene dataset and hundreds of millions of embodied perception and decision pretraining records, without third-party audit or scale comparison against other datasets. Adjudication: mark as a company claim. Source grade: OFFICIAL.
7. Granularity of the customer list. The only named customers are Dreame (vacuums) and Midea (industry partnership); the rest are scenario descriptions (hotel cleaning, restaurant service, indoor precision operation, smart delivery, takeout pickup). Deep cooperation with leading customers in various fields is unnamed. Adjudication: separate named from unnamed. Source grade: OFFICIAL.
8. Headcount. Baidu Baike records nine employees, plainly inconsistent with a Tsinghua PhD team, a stated hiring expansion and the funding raised; it likely reflects an early registry snapshot or an insured-headcount basis. Adjudication: do not adopt the figure; mark headcount as undisclosed. Source grade: DB (doubtful, not adopted).
9. The CNY 7 billion valuation. Convention A, local media (Yangzhou Daily and reprints, MEDIA): "according to available information, Beijing Qianjue Technology ... is currently valued at about CNY 7 billion", a reporter's account with no confirmation from the company or from Gao Haichuan, and the subject is the Beijing headquarters rather than the Yangzhou subsidiary. Convention B, the A+ round announcement and mainstream financial media (OFFICIAL / MEDIA): the specific use of proceeds and the post-money valuation are undisclosed. Adjudication: the report writes it as a media-reported valuation of about CNY 7 billion for the Beijing headquarters, unconfirmed by the company, with the A+ announcement disclosing no valuation. Source grade: MEDIA versus OFFICIAL.
10. Headcount of the Yangzhou entity. Convention A, local media (Yangzhou Daily and the Yangzhou Ecological Technology Innovation City website, MEDIA): more than forty employees in Yangzhou, mostly recruited locally with a management team brought from the headquarters. Convention B, enterprise registry (DB): an insured headcount of zero for Qianjue Technology (Yangzhou) Co., Ltd. and 37 for Beijing Qianjue Technology in 2025. Adjudication: the media figure and the registry figures are recorded side by side and are not converted into each other. Source grade: MEDIA versus DB.
11. The one-billion-item data lines are plans, not capacity. The source wording is entirely in the future tense: the data segment "will build a production line handling one billion data items a year", the company "wants to form a collection and annotation line producing one billion data items a year", and it "will act as a delivery centre, taking on deployment and commissioning tasks for more than fifty robot models". Adjudication: the report records these as planned targets and does not count them as capacity in operation. Source grade: MEDIA.
12. Two meanings of "more than fifty". One sense, the Yangzhou delivery centre taking on deployment and commissioning tasks for more than fifty robot models, a workload count. The other, coverage that the embodied brain "fully adapts to more than fifty robot models across seven major categories including bipedal humanoids and wheeled platforms", a count of product-adapted machine types. Adjudication: task counts and adapted-model counts are stated separately and are not converted into each other. Source grade: MEDIA.
13. Whether a standalone A round exists. Convention A, the Yangtze Evening News and Sina Finance sequence (MEDIA): angel, angel-plus, Pre-A, Pre-A+, Pre-A++ and A+, a cumulative count of nine rounds, with one financing event in September 2026. Convention B, earlier coverage (MEDIA, including a June 2026 article) labels that same September 2026 event an A round and lists provincial state capital such as Shandong New Momentum and Shandong Caijin for it. Adjudication: the named sequence follows convention A; the A-round label is recorded as a naming divergence over one event and no separate A round is created. The phase field of this report stays at series-a, since an A+ round sits inside the A stage and the library records no separate a-plus value. Source grade: MEDIA internal conflict.
Risks and Diligence Gaps
First, the capability-span risk. Most of the hundred thousand devices almost certainly come from vacuums and cleaning products, whose task complexity sits far below the general embodied brain claim. Diligence should request per-scenario device counts, task-type distribution and failure rates.
Second, verifiability of the predictive route. Prediction accuracy and loop reliability under occlusion, delayed feedback and multi-step dependencies have no public benchmark; a held-out comparison against end-to-end VLA baselines is needed.
Third, data walls and the cross-body flywheel. The value of one brain, many bodies depends on customers sharing body data, yet body makers treat data as a commercial asset. Whether an independent model company can actually open cross-body data is a question the company itself leaves open.
Fourth, customer concentration and bargaining power. Named customers cluster around Dreame and Midea, and early investor Zhuichuang Venture is affiliated with Dreame — both a resource and a dependency. If a large customer builds its own brain or switches suppliers, revenue volatility is high.
Fifth, theory-to-product conversion. Polynomial representation, effective degree and causal memory are paper-level tools; engineering maturity and real gains are undisclosed.
Sixth, competition and timeline. Industry consensus puts the breakout at 2028 to 2030 while Qianjue's current revenue rests on low-complexity categories; if general capability arrives slower than expected, the valuation logic reverts from embodied brain to cleaning-appliance algorithm vendor.
Seventh, the diligence checklist. Per-scenario device counts and revenue, per-customer ARPU and renewal rates, external validation results for Polibrain OS, Brain Dock shipments, headcount and R&D share, the Yangzhou data-line and deployment-task targets, the gap between the media headcount of more than forty in Yangzhou and the registry figures, and the media valuation of about CNY 7 billion, which the company has not confirmed, all need priority verification.
Composite Assessment
Technology: 7.5. Predictive world modeling plus brain-inspired partitioning, polynomial representation and a loosely coupled architecture form an internally consistent route, backed by ICML and ICLR-level theory; the CTO's ICLR 2025 workshop outstanding paper is a hard signal. Deductions: no public benchmarks for any key metric and an unproven link from theory to product gain.
Business: 7. Hundred-thousand-scale device connections are rare real volume in the embodied-brain segment, and the Dreame vacuum and Midea partnerships plus hotel, restaurant and delivery coverage show commercialization running ahead of peers; the Yangzhou delivery base and a customer base that media report as more than half Jiangsu-based extend the same picture. Deductions: task complexity does not match the capability claim, high customer concentration, undisclosed revenue and ARPU, and the Yangzhou data lines are plans rather than operating capacity.
Financing: 7. Nine rounds in the several-hundred-million range by the media count, with an A+ round of several hundred million in September 2026 and a Yangzhou delivery base backed by a municipal administration, plus provincial state capital, the Yuanhe system and recurring angel institutions, places capital and industrial resources in place. Deductions: no disclosed post-money valuation, a media figure of about CNY 7 billion carrying no company confirmation, and round naming that diverges across sources.
Product: 7. Four lines covering model, cloud platform, edge compute and data services constitute a complete chain, and Brain Dock's domestic brain-inspired chip support is a differentiator. Deductions: Polibrain OS is still in external validation and Brain Dock shipments are undisclosed.
Market: 7.5. The installed base of cleaning and service robots supplies cash flow that need not wait for a hype cycle, physical-AI and world-model enthusiasm supplies valuation elasticity, and domestic compute adaptation fits the self-controllability agenda. Deductions: a crowded field and a breakout timeline widely pushed past 2028.
Composite: 7.2. Qianjue is this list's hardest-theory, most restrained-narrative company: it tells no humanoid story, instead narrowing the world model into a map for planning and selling that map to robot makers who already ship. Its theoretical work on latent identifiability and effective degree competes on the same question as LeCun's line and came earlier. The physical-AI verdict: the meeting point of world models and physical simulation is state representation — what to predict and what to ignore determines whether a model transfers across scenes, and Qianjue makes that the core. The direction is right; between drawing the map and delivering the lychees, in its own metaphor, lie two unfinished steps: opening cross-body data and long-run validation on real tasks. The Yangzhou base adds a delivery and data-capacity dimension that is still on paper, and the media valuation figure adds an unconfirmed number to the record.
References
- Qianjue official site (qj-robots.com, accessed 2026): perception-decision-control model described as the world's first brain-inspired multimodal large model (ICLR 2023); decomposed decoupled large-organization structure with linear pretraining; cloud development platform with API and web access, a perception model suite and motion-primitive communication; Brain Dock supporting spiking and artificial networks on NVIDIA, AMD, Lynxi and Huawei Ascend; integrated data-model services with in-house annotation
- Sina Finance (2026-10-21): Polibrain released October 20, 2026 at the inaugural Physical AI and World Model Conference in Nanjing; Gao's lychee-and-map analogy contrasting VLA trial-and-error with world-model planning; quadruped trash-collection demo; deliberately not generating future frames, predicting only action-relevant information; the 2025 ICML paper with Tsinghua on Fourier-Walsh analysis and low-order preference; the 2026 Effective Degree proposal with orthogonal polynomial fitting; LeCun's May 2026 LeJEPA paper; counterfactual history filtering; loosely coupled architecture illustrated by the IoT washing machine; nearly 50 adapted bodies
- Zhangtongshe (2026-09-08): several-hundred-million A+ round with Yuanhe Houwang, Xiaoguang, Xinneng, Inno Angel, Jingming and Weilai Bianji; nine cumulative rounds; founded 2023 in Chaoyang, Beijing; core team from Tsinghua's brain-inspired center; the predictive world model line fixed as early as 2017; the no-ChatGPT-moment judgment; first orders in food service, cleaning and hotel robots; CTO Zhang Tianren and Gao's remark that customers care about tidying slippers, not architectures
- Robot Open Community (2026-09): the several-hundred-million A round led by Jingming Capital with Shandong New Momentum, Shandong Caijin, Yuanhe Houwang, Xinneng, Nantong Venture, Inno Angel, Shangshi, Ren'ai and Xuansu; proceeds to world-model architecture, algorithm iteration, deployment and team expansion; brain-inspired partitioning decoupling perception, decision and control; vision-only spatial modeling and planning; adaptation across bipedal, wheeled, quadruped, drone and vacuum forms; scenarios spanning hotel cleaning, restaurant service, indoor precision operation, smart delivery and takeout
- STAR Market Daily (2026-09-09): A+ round including Fengcheng Capital as long-term advisor and participant; the March 2025 Pre-A of tens of millions co-led by Zhuichuang Venture and DT Capital with Zhengjing Zhiyuan; the December 2025 Pre-A++ near RMB 100 million; the one-brain-many-bodies strategy and the company's position in the software and model layer serving robot makers
- Cailianshe and Tencent News (2025-12-22): Pre-A++ near RMB 100 million for R&D, hiring and global commercialization; angel and angel-plus rounds with Inno Angel, TusStar, Shuimu Tsinghua Alumni Seed Fund and Lingyi Venture; three generations of embodied brain pretrained; claims of the world's largest pure real-collected household dataset and hundreds of millions of pretraining records; Polibrain OS entering external validation; deployments in hotel cleaning, restaurant service, humanoid reception and indoor precision operation; expectation of over one hundred thousand devices within 2026
- Daily Science and Technology Network (2025-12): the largest deployment category is semi-service, semi-household cleaning, with Dreame's vacuum line receiving task-scheduling and scene-understanding models; Polibrain OS built on a unified brain-inspired architecture with the unified perception layer engineered
- WeChat account on embodied AI and world models (2026-06): the A round led by Jingming with Inno Angel, Shangshi and Ren'ai re-upping; hundred-thousand-scale device connections; the CTO's paper winning the ICLR 2025 world-model workshop outstanding paper award; the predictive-versus-generative distinction illustrated by a falling cup that only needs to be known as falling and breakable; brain-cerebellum decoupling; the Dreame vacuum case; industry timelines from Pudu's Zhang Tao (three to five years) and Matrix Superintelligence's Zhang Haixing (consumer breakout 2028 to 2030)
- WeChat account on provincial state capital (2026-06): A-round investors including Shandong New Momentum, Shandong Caijin, Yuanhe Houwang, Xinneng and Nantong Venture; deep cooperation with Midea and Dreame; brain-cerebellum decoupling; adaptation across bipedal, wheeled, quadruped, drone and vacuum forms; hundred-thousand-scale devices; proceeds to architecture, algorithms and deployment
- Phoenix and Tencent News (2025-05): Pre-A+ from Junshan Investment, Vertex Ventures and Shixi Capital under GigaDevice, cumulative funding in the several-hundred-million range; the November 2024 angel round of tens of millions led by Inno Angel with Shuimu Tsinghua Alumni Seed Fund, TusStar, Jiushang Capital and Jiadao Gongcheng; NeurIPS Spotlight and TPAMI publications plus a ViZDoom third place; benchmarking Sergey Levine's line seven years earlier; a joint effort with a named 3C industry partner on the world's largest pure real-collected household dataset
- Baidu Baike (accessed 2026): incorporation on June 13, 2023; registry details including legal representative Gao Haichuan and the Chaoyang address; the timeline of October 2024 angel, January 2025 angel-plus, March 2025 Pre-A, May 2025 Pre-A+, December 2025 Pre-A++ and February 2026 extension; a nine-employee figure that conflicts with other sources and is not adopted
- Industry background (public): the two world-model routes of generative pixel reconstruction versus predictive state evolution; VLA and world models converging (DriveVLA-W0 and WorldVLA among 2025 and 2026 work); three tiers of embodied data across internet video, simulation and real hardware, and the difficulty of cross-body reuse
- Yangzhou Ecological Technology Innovation City administrative portal (2026-10-22): the 2026 Physical AI and World Model Conference opened in Nanjing; Qianjue Technology signed with the administrative committee on site and the Qianjue Technology World Model and Robot Brain Yangtze River Delta Base was settled in the Yangzhou Data Industrial Park
- Yangzhou Daily via Jiangsu Provincial Party Committee News (2026-10-22): more than forty employees at the Yangzhou entity, mostly recruited locally with a management team brought from the headquarters; reportedly the Beijing headquarters is currently valued at about CNY 7 billion, worded as a reporter's account without company confirmation
- Yangzhou Net (2026-10-22): report on the signing and the base landing, with the landed entity per enterprise registry being Qianjue Technology (Yangzhou) Co., Ltd., incorporated 2026-07-10, registered capital CNY 30 million, 21/F Tower G of Yangzhou Software Park, legal representative Gao Haichuan, wholly owned by Beijing Qianjue Technology
- Tencent News and Yangtze Evening News via Yangzhou Publish (2026-10-14): future-tense plans for a production line handling one billion data items a year, a collection and annotation line producing one billion items a year, and a delivery centre taking on deployment and commissioning tasks for more than fifty robot models
- Sina Finance (2026-10-22): the embodied brain fully adapts to more than fifty robot models across seven major categories including bipedal humanoids and wheeled platforms
- Beijing Municipal Science and Technology Commission and Zhongguancun Administrative Committee website (2026-10-21): on 20 October 2026 Beijing Qianjue Technology officially released the Polibrain predictive world model, positioned as an embodied-intelligence brain base for the real physical world
- STAR Market Daily, Securities Times and Cailianshe (2026-09-08): the A+ round of several hundred million RMB with Yuanhe Houwang, Xiaoguang Capital, Xinneng Venture, Inno Angel Fund, Jingming Capital, Weilai Bianji and Maple Pledge Fengcheng Capital; the specific use of proceeds and the post-money valuation undisclosed
- Tencent News and Yangtze Evening News (2026-10-19): close to half of the customers come from Jiangsu, with Tiance, Xingong Group and Yika in Nanjing and Dreame, Magic Atom and Pangolin in Suzhou
- Enterprise registry records (DB): insured headcount of zero for Qianjue Technology (Yangzhou) Co., Ltd. and 37 for Beijing Qianjue Technology in 2025
- Local media reports (2026-10): the headquarters has moved south to Nanjing's Xuanwu district, at the Xuzhuang High-Tech Zone and the Jiangsu International Data Port
- Toutiao and investor outlets (2026): the Dreame robot-vacuum business as a named collaboration and the scenarios covered by it
This report is compiled from public sources and does not constitute investment advice. The company is private with no audited financials; device counts, customer names, dataset scale and adapted-body counts are company claims, the hundred-thousand figure appears both as achieved and as a target, the predictive world model has no public benchmark, the A+ announcement discloses no post-money valuation while local media relay an unconfirmed figure of about CNY 7 billion, and the Yangzhou headcount, data-line and deployment-task figures are media and planning statements; uncertainties are flagged throughout. Data is current as of 2 November 2026.
Related papers
Related tools
Related cases
Related reports66
- 51World Deep Dive: A Listed Digital Twin Company Where 51Sim Becomes the First Growth Engine While 51Earth Nearly Vanishes, and Two Irreconcilable Financing Naming Conventions Must Both Be Kept
- AMI Labs: A USD 1.03B Seed Bet on World Models, LeCun's Non-Generative Path and Its Blank Scorecard
- Anshi Pacific (PERA) Deep Dive: A Beijing Domestic-CAE Leader Founded in 2003 Whose PERA SIM General Simulation, AIROM Reduced-Order Models, AgentX Multi-Agent and Jingzhi iGPT/Whale Span from Industrial Simulation to Physics AI
- AspenTech Deep Dive: The Process-Simulation Incumbent Acquired by Emerson and Its Hybrid Industrial-AI Route
- Autodesk Deep Dive: Neural CAD Foundation Models and a $200M World Labs Bet on Physical-World AI, with NAVASTO Adding CFD Surrogate Capability
- Bentley Systems Deep Dive: iTwin Digital Twin and Generative AI Empower Full-Lifecycle Infrastructure Engineering
- BeyondMath Deep Dive: A Foundation Physics AI on the Neural-Operator Route, a $19M Consortium Budget, and Three Misread Facts
- Cadence Deep Dive: System-Level Multiphysics EDA and the Millennium M2000 Supercomputer Anchor Physical AI, with AuraStack Agentic AI and Fidelity GPU CFD as Key Levers
- COMSOL Deep Dive: Four Decades of Multiphysics Solvers, Entering Physical AI via DNN Surrogate-Model Training and All-Physics GPU Acceleration
- Dassault Systèmes Deep Dive: 3DEXPERIENCE and SIMULIA Virtual Twin Physics Behavior Anchor Industrial AI, with LEO Virtual Companion and GPU Explicit Dynamics as Physical-AI Levers
- DP Technology: From the Gordon Bell Prize to National-Level Unicorn Status and the Door of a Hong Kong Listing, the AI-for-Science Leader's Technical Delivery and Commercial Blank
- Emmi AI: From Linz Spin-out to a EUR 330M Mistral Acquisition in 17 Months
- Fainite Deep Dive: A Zurich Physics-AI Simulation Startup Whose Caltech, ETH Zurich, Cambridge and Google Team Uses Physics-Aware Deep Learning to Cut Simulation from Tens of Hours to Seconds, With Only Venture Kick Grants and No Priced Round
- Fysics AI: The PhysX Founder's Second Act, a Domestic Differentiable Physics Engine and Its Ecosystem Play
- Frontop (301313) Deep Dive: From Digital Exhibition Halls to a Physical-AI Data Foundation - FTE Engine, FunSim Simulation and the Gap Between Positioning and Revenue
- Godela Deep Dive: A YC X25 Physics-AI Simulation Startup Whose Ex-Apple, Google and Intel Team Uses Physics-Constrained Surrogate Models to Cut Simulation from Days to Seconds, at Seed Stage with Pilot-Only Access
- HollyWave (688682) Deep Dive: RDSim AI Electromagnetic Simulation and Engineering Agents Close the Simulation-Measurement-Product Loop
- IAT (Alt) Deep Dive: China's First Listed Independent Automotive Design Company Pivots to AI-native R&D with the Yuanfang, Yufeng and Taiyi 3.0 Product System and an NVIDIA Omniverse Partnership, Reporting RMB 975M Revenue and a Widened RMB 250M Net Loss in 2025
- Inductiva Deep Dive: A Closed Case in Open-Solver Orchestration, Eight Premises That Must Be Corrected, and Where Co-Founder and CTO Hugo Penedones Went Next
- Intesim (INTESIM) Deep Dive: A Home-grown Chinese CAE Leader Founded in 2009 with Self-developed Solvers for Structure, Fluid, Electromagnetics, Thermal and Acoustics and Multiphysics Coupling, a ~RMB 100M B Round from CNII and Shenzhen Capital, and 2025 Revenue and Bookings Both above RMB 100M
- iSoftStone (301236) Deep Dive: A ChiNext-Listed IT Services Giant Whose TianShu Platform Builds a Three-Layer Physical-AI Stack with Self-Developed TwinHub Digital Twin, FSim Industrial Simulation, CAE Multiphysics and Phyxis Embodied Control
- Jua Deep Dive: The Earth Physics Transformer as a Large Physics Model for Energy Trading
- Keysight (with ESI Group) Deep Dive: Virtual Prototyping and Hybrid AI Bridge Measurement and Simulation, with Natural-Language Assistants in EDA
- Lightwheel Deep Dive: A Physical AI Data and Evaluation Infrastructure Startup Founded by a Former NVIDIA Simulation Lead, Seven Rounds Totalling Roughly CNY 2.65 Billion at a Valuation Above USD 2 Billion, Seats on Both the Newton and EgoVerse International Standards Committees, With Two Incompatible Accounts of Q1 Orders
- Luminary Cloud: The Physics AI Model Factory Backed by NVentures
- MathWorks Deep Dive: MATLAB/Simulink and Simscape Multiphysics Power Physical AI, with R2026a Copilots and an Agentic Toolkit for Edge Intelligence
- Maniformer (Mifeng) Deep Dive: An Agibot-Spun Physical-AI Data Infrastructure Firm Whose Bodyless MEgo Capture Hardware Is in Mass Production and Which Closed Three Rounds in Half a Year Led by China Telecom
- Mistral AI: How Physics AI Moved From Language Models to EUV Lithography and Nuclear Power
- Monolith AI: The British Engineering-AI Firm That Cut Physical Test Counts Instead of Accelerating Simulation, Now Inside CoreWeave
- Monumo Deep Dive: A Cambridge Deeptech Company Redesigning Electric Motors with the Anser AI Engine, Tens of Millions of Multiphysics Simulations per Day, GBP 10.5M Raised in Total, and a Series A Being Raised
- NAVASTO: Moving Solver Insight Onto the Designer's Desk, and How Autodesk Bought That Entry Point
- Navier: Handing CFD Workflows to Agents, a Seed-Stage Bet on AI for Hardware Engineering
- Neural Concept: An EPFL Spin-out Betting on Neural Surrogates, with $138M Raised
- NP Co. (NP Company) Deep Dive: A 2025-founded Paris Inria-born AI-native simulation startup that raised a EUR 6M pre-seed led by Partech with Mistral AI co-founders as angels, claiming a pre-trained physics foundation model at 1,000x speedup — most claims unverified
- NVIDIA Deep Dive: Building the Full-Stack Physical-AI Infrastructure with PhysicsNeMo, Warp, Cosmos World Foundation Models and Omniverse
- Physical Intelligence Deep Dive: A $2.1B Bet on the GPT Moment for Robotics, With pi0.7 Self-Reporting 60-80 Percent Success on Unseen Tasks and Neither Pricing nor Revenue Disclosed for the Pi API
- Physical Superintelligence Deep Dive: A $58M Seed Bet on Virtual Physicists, With Emmy Targeting AI Data Center Multiphysics Optimization as Its First Commercial Proof Point
- PhysicsX: Pretrained Geometry and Physics for Industrial Engineering at a $2.4B Mark
- Prometheus: A $41B Mark, $18.2B Raised, and Not Yet a Product
- PTC Deep Dive: Creo/Onshape and the ThingWorx Digital Thread Power Physical AI, with NVIDIA Isaac Sim Bridging Robotics Design to Simulation
- Quanscient Deep Dive: Cloud-Native and Quantum-Powered Multiphysics Simulation as the Data Engine for AI-Native Hardware Engineering
- Rescale Deep Dive: An HPC Operating System and Desktop Engineering Workbench in the Cloud, a $115M Series D, and Five Misread Facts
- Schrödinger Deep Dive: Physics-Based FEP+ Simulation and Machine Learning Power Molecular Discovery, with the 2026 Bunsen Agentic Co-Scientist
- Secondmind Deep Dive: A Cambridge Engineering-AI Firm Whose Bayesian Active Learning Cuts CAE Simulation and Test-Bench Runs by up to 90%, with Roughly USD 73M Raised and Mazda Both Customer and Investor
- Shichuang Technology Deep Dive: SupreCAST Cloud-Native Material-Processing Simulation and Industrial Agents Advance the CAE+AI Strategy
- ShonCloud (Shunyun) Deep Dive: Meshless Particle-Method CFD for EV and Nuclear Fluid Multiphysics
- Shuzhi Maijin Deep Dive: A Shenzhen Fully Self-Developed SNN-plus-LBM CFD Startup Claiming Second-Level Results on Hundreds-of-Millions of Grids on Consumer Hardware, With a Tens-of-Millions-RMB Angel Round From a Ministry-of-Finance-Affiliated Fund and an FPGA Migration Plan
- Siemens (incl. Altair) Deep Dive: The 2025 Altair Acquisition Unites Simcenter and Altair, with PhysicsAI Geometric Deep Learning and PhysicsAI Generate as the Physical-AI Core
- Silurian Deep Dive: A Generative Forecasting Transformer Earth-System Foundation Model for Weather Risk on Energy and Infrastructure Assets
- SimScale Deep Dive: A Cloud CAE Platform Built on Core-Hour Subscriptions, EUR 52M of Equity Kept Separate from EUR 25M of Debt, and Five Misread Facts
- Skild AI Deep Dive: A General Robot Foundation Model Incubated at CMU's Robotics Institute, With a USD 1.4B Series C at a USD 14B Valuation, a Conflicting Series B Size, and an ARR Figure That Is a Run-Rate Rather Than Revenue
- Songying Deep Dive: A Huawei-Cloud-Alumni Startup Building a Domestic Physical AI Operating System, With ORCA OS Positioning Against NVIDIA Omniverse, Three Incompatible Financing Narratives, and a Verifiable Third-Party Anchor in a CNY 20 Million Minority Investment
- Sudo Tech Deep Dive: Zero-Real-Data Simulation Training, the Fully In-House Sudo R1 Stack, and an Atomic-Skill Commercialisation Strategy from a Startup Valued Above CNY 20 Billion in 17 Months
- Suochen Technology: The Pain of Becoming a Physics-AI Platform from a CAE Vendor
- Synopsys (Ansys) Deep Dive: The 2025 Ansys Acquisition Turns the EDA Leader into a Silicon-to-Systems Engineering Powerhouse, with Ansys SimAI and the NVIDIA Physical-AI Partnership as the Core Physics-AI Levers
- TenFong Deep Dive: An Academician-Led Chinese Industrial Simulation Software Vendor Whose Official Timeline and Database Records Use Incompatible Round Names, With an Official Financing PDF and an Investment Database Both on File
- Tianfu (Nanjing Tianfu Software) Deep Dive: A 2011-founded Chinese Intelligent-CAE Veteran That Closed a Several-Hundred-Million-RMB Series C, Pivoted to an Industrial World Model in 2024, With Unreconcilable Financing and Valuation Records and No Disclosed Revenue
- Tongyuan Soft Control Deep Dive: MWORKS (Syslab/Sysplorer) as a Domestic MATLAB/Simulink Alternative, with MWORKS AI and the MOGICK Engineering Agent Advancing Home-Grown Physical AI
- Vinci AI: Turning Physics Simulation from a Checkpoint into a Continuous Constraint
- Westlake Robotics Deep Dive: Westlake University's First Robotics Spin-out Builds a Dual-Pretrain Brain-and-Cerebellum Architecture with WR1 Whole-Body Foundation Model and GAE Behaviour Expert
- World Labs: Fei-Fei Li's Spatial-Intelligence Unicorn, Acquired by AMD for $8.2B
- XVERSE Deep Dive: A Simulation-World Route to Physical AI Built on the XVERSE Model Family, an Edge-Cloud 3D Engine and the Physics-Based MotionGen Model
- SimWE: From In-House Solvers to a Physical-AI Engine, the Domestic CAE Leader and What Comes Next
- Yunjing Zhifang Deep Dive: A Peking-University-Incubated Cloud-Native GPU CAE Company with the Simversus Platform and the YJ-DORGP Small-Sample Intelligent Solver, Reporting 275 Percent Order Growth, 160 Percent Revenue Growth and Breakeven in 2025
- Zhongke Diwiji (Five Ages) Deep Dive: A CAS-Institute-of-Automation and Tsinghua Spin-out Whose FAM Embodied Foundation Model Uses Action Optical Flow for Cross-Embodiment Generalisation, Reaching 97% Success from 3 to 5 Demos, with EraBrain and Jiling Robots Serving Sinopec, Leapmotor and Midea
- ZWSOFT (688083) Deep Dive: Overdrive Kernel and ZW3D Wukong 2027 Underpin an Integrated CAD/CAE/CAM Stack, with a CAx Agent Platform as the Practical Route to Engineering AI
This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of the publication date; please refer to the company’s official filings.