Jiutian Artificial Intelligence Platform

What is the Jiu Tian Artificial Intelligence Platform?

The JiuTian Artificial Intelligence Platform is an AI service platform within China Mobile’s ecosystem; it is currently operated by China Mobile JiuTian Artificial Intelligence Technology (Beijing) Co., Ltd. This platform connects computing power, data, models, intelligent agents, and industry-specific applications, offering individual developers, research teams, and enterprises the ability to work with large-scale AI models and to develop such models.

It is not a single chatbot; rather, it is a product ecosystem that includes a large model experience center, a production and development platform, capabilities for aggregating models and agents, as well as industry-specific solutions. Users can first try out the generation and Q&A functions online, and then, depending on their needs, they can integrate APIs, train models, or arrange for enterprise-level deployment.

Product composition and applicable tasks

Main functions

Text generation and natural language interaction

• When users enter questions, writing requirements, or structural constraints, the model can generate answers, outlines, summaries, and text in a specified format; it is suitable for knowledge-based Q&A, document drafting, and information organization.
• The JiuTian basic large model focuses on natural language generation, dialogue, and information extraction; specialized models for fields such as healthcare and government services are designed for specific domains, yet their outputs still need to be reviewed by individuals with the appropriate qualifications or knowledge.
• Multiple rounds of dialogue maintain context, making them suitable for gradually adding conditions and asking follow-up questions; historical conversations, long-term memory, and deletion rules should be used in conjunction with privacy settings.

Image and video generation

• Image generation accepts textual descriptions; reference images of up to 30MB in size can also be used. Users can choose styles such as oil painting, line art, or watercolor, and adjust the aspect ratio of the image.
• The common output formats include square, landscape, and portrait styles, such as 1024×1024, 1125×844, 750×1000, 1125×633, and 750×1333; 1 to 4 images can be generated each time.
• Video generation is used to convert text or visual ideas into dynamic content; the specific models, duration, resolution, waiting time, and quotas vary depending on the capabilities of the account, and the details should be referred to in the user interface.

AI search and document reading

• AI search retrieves and organizes answers around a specific question; users can ask follow-up questions, which makes it useful for quickly obtaining an overview of a topic. However, important conclusions should still be verified by referring to the original materials.
• Document reading is supported for PDF, Word, and Excel files; the maximum size of a single file is 5MB. Features include structure extraction, question-answering based on documents, generation of short summaries, understanding of table data, and locating relevant content.
• Before uploading files that contain personal information, trade secrets, or materials belonging to third parties, it is necessary to verify the relevant authorization requirements, data masking rules, and the organization’s guidelines for data usage.

MoMA Multi-Model and Agent Routing

MoMA is a service for aggregating models and agents, capable of distributing tasks among various models, specialized agents, and complex collaborative processes. Its value lies in allowing users to avoid the need to manually select a model for each task, as well as in enabling the creation of execution plans for complex tasks.

• First, identify the user’s intent, and then determine whether to use a specialized model, a single agent, or a multi-agent process.
• The execution path is chosen based on answer quality, latency, and cost, and the results of multiple models can be optimized.
• It supports custom routing, dynamic feedback learning, and failover; when one option is unavailable, other available paths can be tried.
• Planner can break down complex goals into sub-tasks, then coordinate different agents to carry out those tasks and aggregate the results; it is suitable for multi-step analysis and the automation of business processes.

The number of models and agents listed in the public assistance materials corresponds to a specific version, and should not be regarded as a permanently fixed limit. The platform released MoMA 2.0 in 2026; the actual components available depend on the settings in the current account console.

Model training, fine-tuning, and deployment

• The large-model development platform covers model design, training, fine-tuning, compression, and evaluation, and is suitable for organizations that have their own data and algorithm teams to create customized models.
• The development platform offers capabilities for deployment, hosting, and inference; it allows models to be packaged as callable services that can then be integrated with internal systems, web applications, or business processes.
• Before deployment, companies need to assess the rights related to the training data, the available computing resources, the performance of the models, the responsibilities for maintenance, as well as security and compliance issues. The results obtained from online testing should not be equated directly with the performance in a production environment.

Supported models and industry areas

Tutorial for the JiuTian Artificial Intelligence Platform

1. After logging in to the platform, browse the experience center and help resources first, then register using your phone number and complete the verification required for the account; enterprise features may require additional authentication.
2. Select text dialogue, image generation, video generation, AI search, or document reading based on the task; do not paste sensitive data into unknown environments.
3. In the input field, specify the goal, background, output format, and constraints; for image tasks, choose the reference image, style, aspect ratio, and the number of images to generate.
4. Check the results of the first round, and make further adjustments by asking additional questions, imposing extra constraints, or changing the model; facts, figures, policies, and professional advice require manual verification.
5. When system integration is required, consult the development documentation and the account control panel to check the list of models, tokens, permissions, quotas, and billing details, and then conduct small-scale testing calls.
6. Before going live, the enterprise establishes processes for data masking, access control, log auditing, output review, and error recovery, and defines the scope of services and boundaries of responsibilities through contracts.

API integration and development process

The platform offers various development tools such as model listing queries, Token generation, MoMA calls, online searches, and file uploads. The interface addresses, parameters, available models, and limits may change; developers should refer to the current console and documentation for accurate information.

1. Register and complete the developer or enterprise verification to confirm whether the account has been granted access to the desired models and interfaces.
2. Create or obtain invocation credentials in the account, and store the keys in the server-side key management system; do not include them in website code, public repositories, or client installation packages.
3. First, query the current list of models; then construct texts based on documents, perform searches, handle file-related operations or MoMA requests, and set parameters for timeouts, retries, and concurrency control.
4. Record the number of requests, latency, reasons for failures, and output quality; implement additional verification at the business level for sensitive content, hallucinations, and prompt injection.
5. Before going live, confirm the billing rules, quotas, data processing procedures, commercial licensing, and the written permission to provide services to third parties.

Input, Output, and Restrictions

Prices, free benefits, and refunds

The platform has not yet disclosed a fixed package price that applies to all accounts, a permanent free data allowance, or a uniform price for APIs. The free benefits may change, and the pricing details are subject to those indicated on the purchase page, in the account dashboard, or in the contract.

Unless otherwise specified in the order or contract, the fees that have been paid are generally not refundable. It is necessary to check the model, quota, validity period, overage pricing, invoicing, termination, and refund terms before making a purchase.

APIs, open-source models, and licenses

The fact that the JiuTian Artificial Intelligence platform provides APIs does not mean that the entire platform is open source; likewise, the existence of a public model repository does not imply that all models, data, and services can be used for commercial purposes freely. Developers must examine the code, weights, datasets, and service agreements separately.

Privacy and data security

• Registration and security verification may require a phone number or email address; during use, prompt texts, model outputs, images, audio, videos, files, feedback, search terms, as well as device and log information may also be processed.
• The platform saves conversations in order to maintain context and improve the user experience; the terms of service state that historical sessions are retained for one calendar year from the date they were created. Once a user deletes a conversation, the corresponding entry in the backup may be removed as part of the update process of that backup.
• Interactive data may be used to improve and iterate services after being encrypted and thoroughly de-identified; therefore, confidential information, unpublished code, and sensitive personal data should either have their sensitive elements removed or not be uploaded at all.
• Users can delete conversations and submit requests regarding privacy; once an account is closed or its data is deleted, the platform will remove that data or anonymize it after the shortest retention period required by law has elapsed.
• In principle, information is not transmitted or stored abroad; however, sharing with affiliated companies, authorized partners, or as required by law must be assessed in light of privacy agreements.
• The creator of a personal tool may be able to see the conversations that arise from its tool; therefore, do not submit any information that you do not want the creator to have access to when using tools created by third parties.

Copyright, Content Identification, and Commercial Use

• The platform will add explicit or implicit identifiers to the content generated by AI, and users are not allowed to delete, alter, or circumvent these identifier requirements.
• The agreement permits the user to decide on their own how to use the output, as long as such use is compliant with relevant regulations; however, it does not guarantee that the output will not violate copyright, trademark rights, portrait rights, privacy rights, or other rights.
• Whether the output can be used for commercial purposes and whether the platform can be employed in commercial services are two separate issues; using a platform or API for commercial activities, the development of competitive models, or providing services to third parties usually requires a written agreement or authorization.
• Uploading non-personal content may grant the platform and its affiliates broader rights of use; corporate data and copyrighted materials must have their rights verified before being uploaded.
• High-risk outputs in areas such as medicine, law, and finance can only serve as supplementary information; they cannot replace certified professionals or formal decision-making processes.

Suitable for users and typical use cases

• Individuals and students: experience Chinese large language models, generate text and image content, read short documents, and learn the basics of APIs.
• Developers: Use models, AI search, file processing, or MoMA to integrate generation capabilities into internal tools and prototypes.
• Algorithm team: Responsible for model training, fine-tuning, compression, evaluation, deployment, and management of inference services.
• Educational, medical, governmental, industrial, and financial organizations: Conduct project validation focusing on knowledge-based queries, data processing, process support, and industry-specific models.
• Head of Corporate Technology: Compare in-house development, managed services, and customized solutions, and establish through contracts the requirements regarding computing power, data, service levels, and security responsibilities.

Advantages and capabilities boundaries

Main advantages

• It provides a fairly complete set of capabilities ranging from online experiences and model training to deployment and the development of industry-specific solutions, making it suitable for progressing from prototypes to enterprise-level applications.
• MoMA integrates model selection, agent routing, and complex task planning, thereby reducing the effort required for orchestrating multi-model applications.
• It covers a wide range of languages and industry areas, and provides some open-source models, evaluation tools, and development materials.

Usage restrictions

• The current public page does not list in one place the fixed prices, free usage quotas, concurrency limits, and context length for all models; it is necessary to check each of these details within the account before making a purchase or starting development.
• The model may generate errors, omissions, or non-compliant content; the platform does not provide any absolute guarantee regarding the accuracy, completeness, or continuous availability of its outputs.
• Different terms apply to personal online use, API commercial use, enterprise deployment, and open-source versions; one license type cannot replace another.
• The standalone native client available in the official app store has not been confirmed yet; the availability on mobile devices and the specific access method depend on what the platform offers at the time.

Frequently Asked Questions

Can the JiuTian Artificial Intelligence platform be used for free?

The platform offers public access as well as certain benefits that become available upon account creation, but no permanently fixed free quota has been identified. The number of free uses, the available models, and their validity period may change; the details should be checked on the account page after logging in.

Does the JiuTian Artificial Intelligence platform provide APIs?

Support is provided for various use cases, including tokens, model lists, MoMA, online searching, and file uploading. The actual interface permissions, billing details, concurrency limits, and list of available models must be checked in the console.

Is the JiuTian Artificial Intelligence Platform an open-source project?

The entire platform does not automatically become an open-source product just because some of its warehouses are made public. Models and components such as JIUTIAN-139MoE are indeed available publicly, but the licenses regarding the code, weights, and data vary; for commercial use, it is necessary to check each element individually and obtain the required authorizations.

Can the generated content be used for commercial purposes directly?

The agreement permits users to decide on the subsequent use of the outputs as long as such use is compliant with relevant regulations, but it does not guarantee that those outputs are free from any infringement risks. Using the platform or API for commercial services, providing services to third parties, or developing competitive models may require written cooperation or authorization.

Are uploaded documents and conversations saved?

The platform processes and stores the interaction content to ensure continuous service; the protocol stipulates that historical sessions are retained for one year. Users can delete conversations, but the deletion of backups may take some time, and sensitive information should still be anonymized first.

Can it replace doctors, lawyers, or financial professionals?

No. Models in specialized fields can assist with retrieval, answering questions, and organizing information, but their outputs may be incorrect; they cannot replace the diagnoses of qualified professionals, legal advice, investment judgments, or formal decisions.

Summary

The JiuTian Artificial Intelligence platform is suitable for users who need access to large Chinese-language models, capabilities for multimodal content generation, document understanding, model and agent routing, as well as tools for training and deployment. Individuals can start by using the experience center, while developers and enterprises should check the account permissions, current pricing, data processing rules, commercial licensing terms, and the licenses for various open-source components before integrating this platform into their systems.