DeepOpen
Category: Foundation Models / LLMs
Shenzhen open-source 'new-modality' AI company. It builds the DeepOpen non-autoregressive decision model and spatio-temporal models, and runs the GitCC open-source community. DeepOpen was founded in 2025. The company is led by 袁重桥 (Yuan Zhongqiao). Based in Shenzhen, China. Team size: 11-50. Latest round: Seed. Key investors include Cowin Capital (同创伟业), Hainan Xichen Kechuang Investment (希晨投资).
- Founded
- 2025
- Headquarters
- Shenzhen, China
- Team size
- 11-50
Value proposition
Open-source decision models that return typed outputs (choices, scores, probabilities) that code can consume directly, skipping token generation. The company claims this cuts cost by orders of magnitude versus generative LLMs. It also offers low-cost model hosting and token distribution through a domestic open-source platform.
Products and solutions
DeepOpen decision SDK/engine (Apache-2.0, Laya/ModernBERT/mmBERT-based, Router with english, multilingual and typed-decisions checkpoints), Deepopen spatio-temporal multimodal model platform (text/image-to-video, DramaForge pipeline), GitCC open-source community, aggregated token hub, GCC AI Coding (gitvv.com), DeepMathLab (Scilab plus LLM numerical simulation), 千智 app, enterprise open-source LLM customization and private deployment.
Unique value
One of the earliest open-source alternatives to TypeSafe's closed Jev decision model. Self-reported results are 98.02% on CLINC150 (in-scope only, four-encoder ensemble) and 94.12% on BANKING77. The company claims first place on CLINC150 and second on BANKING77.
Target customer
Chinese AI developers and enterprises needing structured classification and decision automation, private LLM deployment, and low-cost inference tokens.
Industries served
Customer service and intent routing, banking and fintech, embodied AI and robotics, film and short-drama generation, enterprise software and developer tools
Technology advantage
Non-autoregressive System-1 decision engine with calibrated probabilities, built on the RLCD calibrated-decision reinforcement learning framework. Backed by the Trustworthy LLM Lab run jointly with Peking University Shenzhen Graduate School. The company claims the GitCC community has over 1M users and over 1M open-source projects, and that its token costs are below 10% of comparable overseas platforms. The GitHub repo has about 3.1k stars and 343 forks.
How they differentiate
Fully open-source and locally runnable, targeting cheap intent classification on public benchmarks. It is bundled with a domestic open-source hosting community (GitCC) and a token-distribution platform, and is positioned as Chinese sovereign open-source infrastructure.
Main competitors
TypeSafe (Jev), StartLux / 原点星辉 (StartLux-Decision), AutoTrust AI Lab (JEV-27B-VL)
Key partnerships
Peking University Shenzhen Graduate School (joint Trustworthy LLM Lab), Shenzhen Software Industry Association
Major milestones
2025-09: company founded in Shenzhen (Sep 5). 2025-11: ¥10M+ seed round from Cowin Capital (announced 2026-01-06). 2026-09: second round of tens of millions of RMB from 希晨投资. 2026-10: open-sourced the DeepOpen decision model, which reached the GitHub daily trending list, the joint PKU Shenzhen Trustworthy LLM Lab began operations.
Growth metrics
Company-reported: model weekly downloads above 10,000; GitHub repo about 3.1k stars and 343 forks; GitCC community above 1M users; core AI training platform downloads above 100k; token distribution grew tens of times over in recent months.
Market positioning
Early-stage Chinese challenger in the emerging 'decision model' category triggered by TypeSafe's Jev. The company also operates a domestic open-source community and AI infrastructure.
Geographic focus
China (Shenzhen / Greater Bay Area), with global open-source distribution via GitHub and HuggingFace
Patents and IP
Open-source code (Apache-2.0) on GitHub and HuggingFace with an accompanying paper. The lab cites 'SeLaR: Selective Latent Reasoning in LLMs' (ACL 2026 SAC Highlights). No patents found.
About 袁重桥 (Yuan Zhongqiao)
Ex-Sugon (中科曙光) supercomputer R&D; Peking University BS/MS. National Mathematical Olympiad first prize; co-initiator of the PKU (Public Knowledge Unit Agreement) open-source license.
Latest news about DeepOpen
More Foundation Models / LLMs companies
Official website: https://www.deepopen.com