Qingyan Series A

Qingyan Tech Deep Dive: A Geometric-Physics-Engine Company Incubated from Tsinghua and Yanqi Lake Institute, Building a World Model from Differential Geometry with a Model-Data Resonance Flywheel, Closing a Nine-Figure A Round in August 2026

Qingyan Tech was incubated from Tsinghua and the Beijing Yanqi Lake Institute of Applied Mathematics, builds a geometric physics engine from differential-geometry theory toward a world model, proposes a model-data resonance flywheel, a 4D world-state data centre and Neo Lab, and closed a nine-figure Series A led by Legend Capital in August 2026.

Executive Summary

Qingyan Tech (Beijing Qingyan Technology Co., Ltd.) is a physical-AI company incubated from Tsinghua University and the Beijing Yanqi Lake Institute of Applied Mathematics. Its technical thesis is to build a geometric physics engine grounded in differential-geometry theory and to use that engine as the route toward a world model, rather than relying solely on data-driven learning. The company articulates a "model-data resonance" flywheel in which the geometric model and collected data reinforce each other, operates a 4D world-state data collection centre, and runs a "Neo Lab" research unit. It closed a nine-figure Series A in August 2026 led by Legend Capital, with Suzhou Ventures and Ningbo Yishi participating, and holds Beijing municipal specialised-and-innovative and national high-tech enterprise status.

Company Overview

Qingyan Tech was established in November 2022 and is registered in Beijing. Its origin story is academic: it was incubated from Tsinghua University and the Beijing Yanqi Lake Institute of Applied Mathematics, which signals a mathematics-first founding rather than a pure product start-up. The chairman is Sun Mingming and the chief executive is Zhang Yingwei. The company holds Beijing municipal specialised-and-innovative (little giant-track) recognition and national high-tech enterprise status, which are useful credibility markers for a Chinese deeptech firm.

Technology

The distinctive claim is the geometric physics engine built on differential-geometry theory. Where most world-model and embodied-AI companies learn dynamics from data, Qingyan argues that a geometry-grounded engine provides physical priors that data alone cannot supply, and that this engine is the path to a dependable world model. The "model-data resonance" flywheel describes a loop in which the geometric model generates structure and the 4D world-state data centre supplies observations, each improving the other. The 4D qualifier suggests spatial plus temporal state capture for physical scenes. Neo Lab is the internal research unit responsible for advancing the engine and the world-model agenda.

The strength of this approach is a clear theoretical differentiation and a longer operating history than the 2026 pure-startups in this list. The weakness is verification: the geometric-engine claims and the resonance flywheel are described conceptually, with limited public benchmarking against data-driven baselines.

Product and Commercial Model

Qingyan's product surface combines an engine, a data centre and a research lab rather than a single shipped product. The intended customers are embodied-AI developers and industrial-simulation users who need physically grounded world models and data. No explicit pricing or revenue has been disclosed. The commercial logic is to own the geometric substrate and the data loop, then monetise through licensing, data services or platform access, though the precise model is not yet public.

Competitive Landscape

Qingyan competes with the broader world-model and embodied-AI field and with geometric-deep-learning researchers, while differentiating on a mathematics-first, geometry-grounded engine. Its longer history and academic incubation give it more credibility than a fresh 2026 entrant, but it remains smaller in profile and funding than the best-capitalised peers such as Physical Intelligence, Skild AI and the Chinese embodied leaders. The geometric-engine thesis also competes with the dominant data-driven paradigm, which currently has far more deployed evidence.

Financing and Capital Structure

The Series A in August 2026 was a nine-figure round led by Legend Capital, with Suzhou Ventures and Ningbo Yishi participating. A nine-figure A after a 2022 founding and municipal specialised-and-innovative recognition indicates steady rather than explosive financing, consistent with a mathematics-incubated deeptech path. No valuation was disclosed.

DISCREPANCIES

1. Establishment and incubation. The company was established in November 2022, while its incubation from Tsinghua and the Yanqi Lake Institute is described narratively; the precise relationship between the institute and the legal entity is not documented. Source grade: OFFICIAL versus MEDIA.

2. Funding amount. The Series A is described as nine-figure renminbi led by Legend Capital with Suzhou Ventures and Ningbo Yishi; the exact figure is not disclosed. Source grade: MEDIA.

3. Geometric-engine distinctiveness. The differential-geometry grounding is presented as a key differentiator, but its advantage over data-driven baselines is described conceptually without a public benchmark. Source grade: OFFICIAL, unverified.

4. Model-data resonance definition. The flywheel is described without a public, precise specification of how model and data reinforce each other measurably. Source grade: MEDIA.

5. 4D world-state centre scale. The data collection centre is described as 4D without a disclosed scale, sensor suite or throughput. Source grade: OFFICIAL, unverified.

6. Recognition status. The company holds Beijing municipal specialised-and-innovative and national high-tech status, which are confirmed credentials but distinct from commercial traction. Source grade: OFFICIAL.

Risks and Diligence Gaps

The primary gap is verification of the geometric-engine thesis. The differential-geometry grounding and the model-data resonance flywheel are conceptually attractive but lack public benchmarking against the dominant data-driven world models. A second gap is commercial evidence: no pricing, no disclosed revenue and no named customer beyond the research-and-data posture. A third gap is scale: Qingyan's profile and funding are smaller than the best-capitalised peers, which matters in a capital-intensive race. A fourth gap is the paradigm bet itself, because geometry-grounded engines must prove they generalise and ship faster than the data-driven methods that currently dominate deployed systems.

Overall Assessment

Technology scores 7.0. The differential-geometry grounding and the longer operating history are real differentiators, and the academic incubation provides mathematical depth. Deducted points: no public benchmark against data-driven baselines, resonance flywheel under-specified, verification gaps.

Commercialisation scores 6.0. Municipal specialised-and-innovative and national high-tech recognition are credible markers, and the Series A shows steady progress, but there is no disclosed pricing or revenue. Deducted points: no customer or revenue disclosed, commercial model not yet public.

Financing scores 6.5. A nine-figure Series A led by Legend Capital is a solid outcome for a mathematics-incubated deeptech, though smaller than the largest peers. Deducted points: exact amount undisclosed, no valuation disclosed, slower pace than explosive seed rounds.

Product maturity scores 6.5. The engine, data centre and Neo Lab are described with concrete structure, more so than pure research announcements, yet no single commercial product is clearly shipped. Deducted points: research-and-data posture, no customer reference, packaging evolving.

Market position scores 6.5. The geometric-differentiation is distinctive and the academic origin is a moat of sorts, but the company is smaller in profile and funding than the field leaders. Deducted points: smaller scale, paradigm bet against the dominant data-driven approach, limited public visibility.

The composite view is 6.5. Qingyan Tech is a mathematics-first, geometry-grounded entry into the world-model race, with a credible incubation story and steady financing, but with the standard deeptech deficits of unverified thesis, no disclosed revenue and a smaller profile than the best-funded peers. The most informative next disclosures will be a public benchmark of the geometric engine against data-driven world models, a first commercial deployment, and clarity on how the model-data resonance flywheel is measured.

References

  1. Qingyan Tech official materials and coverage, 2026: geometric physics engine grounded in differential geometry as the route to a world model; model-data resonance flywheel; 4D world-state data collection centre; Neo Lab.
  2. Company founding and incubation coverage, 2022 to 2026: establishment in November 2022; incubation from Tsinghua University and the Beijing Yanqi Lake Institute of Applied Mathematics; chairman Sun Mingming, CEO Zhang Yingwei.
  3. Series A coverage, August 2026: nine-figure round led by Legend Capital with Suzhou Ventures and Ningbo Yishi.
  4. Recognition coverage, 2026: Beijing municipal specialised-and-innovative status and national high-tech enterprise status.
  5. Analysis of geometric-deep-learning and differential-geometry approaches to physical simulation, 2026.
  6. Analysis of the world-model and embodied-AI field, 2026: comparison with Physical Intelligence, Skild AI and Chinese embodied leaders.
  7. Coverage of data-centre and data-loop strategies for physical-AI training, 2026.
  8. Startup databases and company filings, 2026: registration date, investor roster and Series A status.
This report is compiled from public sources and does not constitute investment advice. The company is private with no audited financials; the funding amount, the differential-geometry advantage, the model-data resonance flywheel and the 4D data-centre scale are drawn from media coverage and company statements with no public benchmark, and no revenue has been disclosed; uncertainties are flagged throughout. Data is current as of 5 November 2026.

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.