Deep Principle
Generative models for chemical reactions and materials (OA-ReactDiff, React-OT), an agent called Mira, and a high-throughput lab, sold to chemicals and consumer-goods companies.
Money and evidence
Latest round, Series A3, 2026-07-14: the record then held 3 research entries, highest level 4. The first research entry is dated 2025-06-24; the highest level so far is 4 (Tested in the physical world).
Rounds and research, on one timeline
- 2026-07-14Series A3Round
undisclosed
For: its AI Scientist, lab pipelines and partnerships
Models and platformWet labCommercial roll-out
'Nearly 1bn yuan' is the cumulative total across the A series, not this round.
Lead · Fortera CapitalT-CapitalKangjun CapitalShunxi FundGL VenturesVertex Ventures China ✓Gobi ChinaExisting investor · Linear Capital ✓Existing investor · Legend Star ✓Existing investor · Baidu Ventures ✓Existing investor · Inspiration Capital
- 2026-03-06Series A2Round
undisclosed
For: LLM + diffusion models, the Mira product line, its lab and in-house pipeline
Models and platformCommercial roll-outWet lab
Announced by XtalPi, which calls Deep Principle a company it incubated.
Lead · Jinma InvestmentXtalPiInspiration CapitalBaidu Ventures ✓
By this date the record held 3 research entries, highest level 4
- 2025-12-25Research
Level 4 · Tested in the physical world
SAGA, an agent that rewrites its own design goals, gets one E. coli hit from 28 tested molecules and three PD-L1 nanobody binders from 24
The paper's claim is that design agents fall short at choosing what to optimise, so the agent should rewrite its targets as it goes. The lab results are first hits from the authors' own assays: the antibiotic screen used a helper that opens the bacterial outer membrane, and one weak hit is far from a drug. The materials and chemical-process designs were checked only by calculation.
- 2025-12-17Research
Level 2 · Computed
SDE, a 1,125-question discovery benchmark from Deep Principle and 25 institutions: the best models score 0.60-0.75, against 0.86 on GPQA-Diamond
Exam-style benchmarks put these models near the top on science questions; put a question back in a researcher's context and a quarter to two-fifths of it goes wrong. More thinking time mostly does not close that, which points at what the models lack rather than how long they get. The scores were computed by the team that assembled the questions, and the company that led the work sells scientific agents. The paper declares no competing interests.
- 2025-11-24Series ARound
over 100M yuan
For: its Mira agent, an L4 high-throughput autonomous lab, and customer work
Models and platformAutomated labCommercial roll-out
Co-lead · Alibaba Entrepreneurs Fund Greater Bay Area fund (managed by Gobi Partners)Co-lead · Ant GroupLenovo Capital ✓Taihill Venture ✓Baidu Ventures ✓
By this date the record held 1 research entry, highest level 2
- 2025-06-24Research
Level 2 · Computed
AIMNet2-rxn, a machine-learned potential trained on reacting molecules, puts organic reaction barriers within about 2 kcal/mol of the quantum chemistry it learned from
General-purpose machine-learned potentials are trained mostly on stable molecules, and the authors show two of them missing barriers by 15 to 25 kcal/mol on a standard reaction benchmark, too far off for kinetics. Training on reacting structures closes most of that gap, at a speed that makes searching millions of pathways practical. Every check is against DFT, the model's own teacher, including an 11-step glucose pyrolysis pathway taken from an earlier computational study that ends at the product experiments observe. Charged, open-shell and metal-containing reactions are outside its scope.
- 2025-03Pre-ARound
on the order of 100M yuan
For: rolling out its ReactiveAI platform in materials, energy and pharma
Commercial roll-out
No lead named; Mingde Capital was sole financial adviser.
Lenovo Capital ✓Legend Star ✓Jinqiu Capital ✓Baidu Ventures ✓Inspiration CapitalExisting investor · GL VenturesExisting investor · Linear Capital ✓
On the deal: Zhiqiang He (Senior Vice President of Lenovo Group; President of Lenovo Capital, Lenovo Capital)
- 2024-11-05Seed++Round
undisclosed
For: product R&D, commercial markets and international expansion
Models and platformCommercial roll-out
- 2024-10-09Seed+Round
undisclosed
For: R&D, team and partnerships with chemicals and materials firms
Models and platformHiringCommercial roll-out
- 2024-06-06SeedRound
nearly $10M
For: R&D, hiring and market expansion
Models and platformHiringCommercial roll-out
XtalPi and DP Technology joined as industry investors.
Lead · Linear Capital ✓Verity VenturesTaihill Venture ✓XtalPiDP Technology
On the deal: Yingzhe Zeng (Partner, Linear Capital)
The investor's own page says · Linear Capital: 2024
Research entries are the timeline's graded entries that name this company. A level grades one claim, not the company, and a blank can mean the record has not caught up rather than that nothing was published.
On the investors' own sites
Investors that name this company on their own site, in a portfolio or their own announcement. A ✓ in a round marks the same investor.
- Baidu Venturesbaidu-ventures.com ↗
- Linear Capital2024linear.vc ↗
- Taihill Venturetaihill.vc ↗
- Vertex Ventures Chinavertexventures.cn ↗
- Legend Starlegendstar.com.cn ↗
- Lenovo Capitalcapital.lenovo.com ↗
- Jinqiu Capitaljinqiucapital.com ↗
Its investors also backed
Other companies on this page whose rounds name the same investor.
- GL VenturesHelixon (Earendil Labs), StoneWise, Zelixir, DP Technology
- Baidu VenturesMoleculeMind, Insilico Medicine, DP Technology
- XtalPiMETiS TechBio, Rhinovate
- Lenovo CapitalMoleculeMind, DP Technology
- Linear CapitalStoneWise
- Legend StarZhichong Pharma
- Shunxi FundRhinovate