OpenBMB
OpenBMB (Open Lab for Big Model Base) is an open-source AI research organization/community that aims to build foundation models and systems toward AGI. Its stated vision is “Big Models for Everyone”; it focuses on end-cloud collaboration, low-power inference, multimodal perception, and full-stack open-source solutions from model training through edge deployment.
At a Glance
- AI and machine-learning developers
- Smart-hardware and embedded-device manufacturers
- Mobile and edge-computing developers
- Researchers and academic labs
- +2 more
AI Tools by OpenBMB
(5)
UltraRAG
MCP Based RAG Framework
VoxCPM
Open Source Text to Speech Model
AgentVerse
Open Source Multi Agent LLM Framework
MiniCPM
Efficient Edge Device LLMs
ChatDev
Multi-Agent AI Software Development
Discussions
No discussions yet
Be the first to start a discussion about OpenBMB
Latest News
ModelBest and OpenBMB announced MiniCPM-V 4.7, a 35B-A3B/1B on-device multimodal release for phones, tablets, and edge devices.
OpenBMB published a technical article on self-replay fine-tuning for extending model capability while reducing catastrophic forgetting.
MiniCPM5-2B was released as a compact 2B-class dense model for on-device and resource-constrained use.
OpenBMB announced its first group of global MiniCPM ambassadors and published a small-model cost/performance optimization initiative.
Products & Services
Compact open-weight language-model family designed for efficient inference and deployment on resource-constrained and edge devices; later generations include MiniCPM4, MiniCPM4.1, MiniCPM5, and MiniCPM-SALA.
Compact 1B-class dense on-device model with deployment and fine-tuning Agent Skills and cookbook support across common backends.
Compact 2B-class dense model for on-device and resource-constrained use, paired with deployment and fine-tuning Agent Skills.
1.3B multimodal model for text, image, and video input with text output, emphasizing visual-token compression, efficient inference, and edge deployment.
Market Position
OpenBMB positions itself around compact, capable, open-weight foundation models optimized for edge/on-device deployment and multimodal interaction rather than only large datacenter models. In the sub-2B segment, external analysis compares MiniCPM-V 4.6 with Qwen3.5 and other small open-weight models; the broader competitive set includes Qwen, Llama, Gemma, Phi, and other open model families. Its differentiation is the combination of small-model efficiency, multimodality, permissive open releases, and deployment tooling.
Leadership
Founders
Tsinghua University Natural Language Processing Laboratory (THUNLP)
Tsinghua University's NLP research laboratory; identified by OpenBMB as an initiating council unit and the lab with which the organization is affiliated.
ModelBest Inc. (面壁智能)
Beijing AI/model company focused on large-model technology and on-device AI; identified by OpenBMB as an initiating council unit and by third-party descriptions as its backing organization.
Executive Team
Zhiyuan Liu
Research leader / Tsinghua University professor associated with OpenBMB
Tsinghua University computer-science professor and NLP researcher; reporting describes him as the academic leader behind the lab/company collaboration and OpenBMB vision.
Board of Directors
Founding Story
OpenBMB is described by Artificial Analysis as a China-based lab jointly founded in 2022 by Tsinghua University's NLP Lab and ModelBest Inc. OpenBMB's own site identifies those two organizations as its initiating council units and frames the effort around making large models accessible to households, developers, and hardware makers.
Business Model
Revenue Model
Public materials describe an open-source research and model ecosystem: models, code, datasets, and agent frameworks are distributed through OpenBMB's website, GitHub, Hugging Face, and ModelScope.
Target Markets
- AI and machine-learning developers
- Smart-hardware and embedded-device manufacturers
- Mobile and edge-computing developers
- Researchers and academic labs
- Enterprises needing private or low-cost local inference
- Open-source AI communities and model integrators
- On-device assistants and local/private AI
- Mobile and edge multimodal understanding
- OCR, document and chart interpretation, STEM problem solving, and long-video understanding
- Voice generation, voice cloning, and multilingual/dialect speech applications
- Software development and consulting multi-agent systems
- Game environments and social-behavior research with simulated agents
- Huawei Cloud
- MediaTek
- Intel
- Baidu Intelligent Cloud