The Fudan University team independently developed and open-sourced it.MOSS became the first in ChinaConversational large-scale language modelsIt combinesPowerful natural language understanding and communication capabilities, diverse application scenarios, and top-notch data security.MOSS has set a new benchmark in the field of AI training models. With its multi-turn dialogue, code generation, and fact-based question answering features, it offers a wide range of applications for academia and industry. Its open-source nature and Chinese optimization give it a unique advantage in local AI applications.
MOSS: A new benchmark for domestically developed conversational AI training models created by a Fudan University team.
Amidst the burgeoning era of artificial intelligence, China has achieved a major breakthrough in indigenous AI technology. MOSS, developed by...The Fudan University team independently developed and pioneered open-source the technology.MOSS became the first large-scale conversational language model released in China, attracting global attention from the technology community.Powerful natural language understanding and communication capabilities, diverse application scenarios, and top-notch data security.It has already set a new benchmark in the field of AI training models.

MOSS official website:https://moss.fastnlp.top/
MOSS main functions
MOSS (Mind Open Source System) is a software application that integrates multiple capabilities such as natural language processing, knowledge-based question answering, and content creation assistance.Open source AI training model platformSince its public beta launch in 2023 and subsequent open-source release, MOSS has brought a vast space for AI applications to academia and industry with its features such as multi-turn dialogue, code generation, and factual question answering.
Main functions overview
| Functional modules | Detailed information | Applicable Scenarios |
|---|---|---|
| Multiple rounds of dialogue | It supports long context, emotion understanding, and logical reasoning, simulating real human-to-human communication. | AI assistant, online customer service, educational support |
| Programming generation and debugging | It can automatically generate code snippets and scripts, supports multiple languages such as Python, and can also provide optimization suggestions based on program requirements. | Developer assistance, teaching, and automated office tools |
| Facts Q&A | By combining big data from the internet and encyclopedic knowledge, we provide users with reliable and detailed answers. | Intelligent search, news fact verification, online knowledge base |
| Creative assistance | Copywriting, article continuation, and style variation enhance the creativity and quality of text content. | Content creation, self-media operation, marketing promotion |
| Multiple types of role-playing | Supports the creation of characters from ancient and modern times, both Chinese and foreign, and allows for dialogue in different contexts. | Educational coaching, cultural and recreational activities, and science popularization consultation |
| Instant AI Search | Integrated intelligent search, optimizing search results in real time based on user conversation content. | Efficiently find information, conduct research and analysis, and support decision-making. |
According to the project leader, MOSS is based onConversational Large Language Model Technology StackIt has opened up all the key paths to "understanding human intent + diverse output". Its independently developed Chinese recognition and generation architecture makes it outstanding in handling complex semantics, comparable to leading international AI products. For example, MOSS can not only complete basic question-and-answer interactions, but also...Multi-turn context, cross-domain reasoningIt has excellent adaptability.

MOSS Pricing & Solutions
MOSS belongs toOpen source AI modelsIt is completely free for academic, educational, and personal users.Official platformUsers can directly experience the online version of the dialogue model without registration or payment, which improves the accessibility and convenience of AI tools and services.
| Product Solution | Is it open source? | Use object | Access methods | Fee details |
|---|---|---|---|---|
| MOSS Online Experience | 是 | Researchers, businesses, developers, and individuals | moss.fastnlp.top | free |
| MOSS offline deployment | 是 | Enterprise-level customization | Download the source code to build your own environment or deploy in collaboration. | According to the agreement in the contract |
It is worth mentioning that the team alsoGitHubThe source code has been opened up, allowing anyone to participate in model training, customization, and redevelopment, which has greatly promoted the openness and sharing of China's AI training model ecosystem.
How to use MOSS
MOSS has an extremely low barrier to entry. Users can access it via a web browser or set it up locally to experience a wealth of AI features such as dialogue, question answering, and content creation. The following is a typical usage flow:
Beginner's Guide

- Visit the official website:Click to experience
- On the homepage, select the "Dialogue/Q&A/Code" function and directly enter your question or text requirements.
- Get AI-generated responses in real time and support continuous interaction in multi-round conversations.
- For the code generation function, you can input a requirement description, and MOSS will automatically output standardized code content.
Offline deployment (suitable for enterprises and developers)
- Go toGitHub project pageCloning source code
- Follow the official documentation to complete local model loading and environment configuration.
- Supports local AI training model customization for large-scale text, datasets, and code.
| step | Operating details |
|---|---|
| 1 | Download the latest source code |
| 2 | Install dependencies |
| 3 | Obtain and load the pre-trained model |
| 4 | Start the service (REST API/web interface, etc.) |
For MOSS user help and development documentation, please visit [website address].GitHub Open Source Homepage】
MOSS is suitable for:
MOSS's ease of use and powerful functionality make it widely adaptable to the needs of users in different fields:
Main user profile
- Academic researchers:Conduct NLP and AI algorithm research, and build customized AI training model experiments.
- Developers/Programming Enthusiasts:Automatic code generation and integration with development tools improve code output efficiency.
- Educational institutions/teachers/students:Assistance in explaining complex knowledge points and provision of writing and Q&A services.
- Businesses/Content Creators:Automated copywriting, data retrieval, and improved output and promotion efficiency.
- Regular users:Intelligent companionship, daily knowledge Q&A, and entertainment interaction

| industry | Typical application cases |
|---|---|
| Educational Research | Chinese/English composition generation and problem-solving explanation |
| IT Development | Automatic code generation and script optimization |
| Media and self-media | Quickly generate promotional copy and creative short articles |
| Business Consulting | Smart assistant, automated customer service response |
| Personal learning/entertainment | Celebrity dialogues, AI interactive chat, knowledge retrieval |
MOSS's multilingual support and open-source nature allowThe broad academic and industrial communitiesUsers can enjoy the benefits of domestically produced AI training models with zero barriers.
Technological advantages and innovation
What did MOSS do right?
- Open source:This is the first domestically developed conversational large language model, with completely open-source code, training data, and model parameters, promoting industry standardization.
- Chinese contextual depth optimization:Optimized specifically for Chinese and multilingual environments, supporting in-depth cross-domain and multi-topic dialogues.
- Security and Privacy:Adopting a locally controlled deployment eliminates the risk of data leakage and facilitates sensitive enterprise applications.
- Strong adaptability:It can be fine-tuned and incrementally trained as needed, and supports secondary development and business integration.
Official Certifications & Open Source Status
| project | state |
|---|---|
| Model open source | √ |
| Training dataset open source | √ |
| Github community support | √ |
| Document completeness | good |
| Enterprise secondary development authorization | support |
| International Application Expansion | It has been promoted globally. |

For more information on the MOSS model architecture, latest papers, and technical reports, please visit:Fudan NLP Lab Homepage和MOSS official release page
Frequently Asked Questions
1. How does MOSS differ from other AI training models such as ChatGPT and ERNIE?
MOSS is China's first open-source Chinese conversational language model, independently developed by Fudan University, and designed to meet diverse local needs.It is compatible with the innovation mechanisms of the open source community. Compared with international/domestic models such as OpenAI's GPT and Baidu's ERNIE, it emphasizes the advantages of open ecosystem, domestic independence, and flexible offline deployment, and supports customized development.
2. How does MOSS ensure data security and privacy?
MOSS supports local private deployment, where all data and applications run entirely under user control. Whether it's text processing, question answering, or custom AI training models, there's no need to worry about data leaving the country or being stored by a third party, making it suitable for enterprise or organization application scenarios with high security and compliance requirements.
3. How can ordinary users participate in the MOSS project or provide feedback?
Regular users can directlyOnline experience platformUsing MOSS is also welcome.GitHub open source project pageParticipate in code contributions, model training, and feature suggestions. The MOSS team also welcomes collaboration from academia and industry to jointly promote the development of the domestic AI training model ecosystem.
MOSS's openness and innovation have injected strong momentum into China's AI industry. As a pioneer in the independent research and development and industrial application of domestically produced AI training models, MOSS not only boasts leading technology but also takes it upon itself to be open to all industries without barriers, becoming an important bridge for promoting the Sinicization and internationalization of artificial intelligence. With the continuous iteration of the AI platform and the addition of community forces, the future of MOSS deserves the continued attention of every AI practitioner.
data statistics
Data evaluation
This site's AI-powered navigation is provided by Miao.MOSSAll external links originate from the internet, and their accuracy and completeness are not guaranteed. Furthermore, AI Miao Navigation does not have actual control over the content of these external links. As of 2:27 PM on August 5, 2025, the content on this webpage was compliant and legal. If any content on the webpage becomes illegal in the future, you can directly contact the website administrator for deletion. AI Miao Navigation assumes no responsibility.

