Ant Group's LingGuang Tops One Million Downloads in Four Days as AI App Building Takes Off

A fast launch puts app creation in the spotlight

Ant Group's LingGuang passed one million downloads within four days of its launch in China, according to the company, then exceeded two million by its sixth day. As of November 24, 2025, the new artificial intelligence assistant had also climbed to sixth place among all free apps on Apple's mainland China App Store. Its main attraction is a feature that turns a sentence into a working mini app in about 30 seconds.

The rush to try that feature, called Flash App, put pressure on the service almost immediately. Three days after launch, on Thursday evening, Ant temporarily suspended it during peak traffic while engineers added server capacity. The feature returned that same evening, and the company carried out further rounds of emergency expansion as demand continued.

Ant says LingGuang reached its first million downloads faster than ChatGPT and Sora did after their initial releases. By Saturday, it had taken first place among free utility apps in Apple's mainland China store and ranked seventh across all free apps. It rose to sixth early the following week.

LingGuang is available through the Apple App Store, major Android stores and the web, for users both inside and outside China. Its early appeal rests on a simple promise: people who have an idea for a calculator, planner, game or other small tool can try building it without first learning a programming language or hiring a developer.

The download surge also exposed the work behind that promise. Each request requires the service to produce an app, and users often return with changes. Ant says a typical session involves about six rounds of edits, making the ability to revise a first attempt a central part of the experience.

What a sentence can turn into

Flash App accepts a plain language request, such as creating a weekly fitness planner or a recipe generator for meals low in sodium. About 30 seconds later, according to Ant, it returns a functioning mini app with buttons, input fields and the logic needed to make those controls work.

A user can then ask for changes through the same conversation. That might mean adjusting the layout, adding a rule or connecting a data source. The assistant keeps track of the latest version, and the resulting app can be shared with friends through a link or QR code. Where suitable, it can connect to backend services, such as spreadsheets or simple databases. These are the systems behind an app that store information or help it carry out tasks.

Ant has highlighted early examples including a generator of activities for children, a calculator for car cost savings and tools for training new employees. These uses combine small, familiar tasks with software that can be adjusted to a particular family, classroom or workplace.

He Zhengyu, Ant Group's chief technology officer who oversees the project, described the intended role of the assistant:

LingGuang is bringing every user their own personal AI developer, someone who can code, create visuals, build programs, and turn complex ideas into simple solutions, right in your pocket.

The conversation continues after the first app appears. Ant says some early users made more than 100 changes to their apps in a short period. The average of about six rounds of revisions per session suggests that people commonly use the first version as a starting point, then adjust it until it better fits their needs.

How vibe coding changes the workflow

Vibe coding describes a way of building software by telling an AI what the software should do. Instead of writing the instructions in a programming language, a person describes the goal in everyday words and asks the system to generate the code.

Many AI coding tools suggest or write pieces of code that a developer still needs to review and combine with other parts of a program. Tools such as Replit or Copilot help developers and hobbyists generate code that is then compiled, tested and deployed. In simple terms, those steps turn the code into something that runs, check its behavior and make it available to users.

LingGuang takes the process further by generating a working mini app that runs immediately, within the chat experience and on the phone. It then stays in the conversation to revise that app. Removing several steps between an idea and a usable result opens the process to people who have never used a code editor or opened a code repository.

The commercial market for this approach is also growing. Replit, based in the United States and a leader in vibe coding, reported annual recurring revenue of about $100 million in the middle of 2025, compared with around $10 million at the end of 2024. Annual recurring revenue is a measure of revenue from continuing subscriptions expressed as a yearly amount. Those figures represent roughly a tenfold increase.

LingGuang's early usage points to the importance of revision within this workflow. A first request may produce a useful draft, while later requests help specify the rules, appearance and information the user actually needs. The app's ability to keep that process inside a conversation is part of its appeal to people without programming experience.

Answers that use pictures, movement and data

LingGuang is presented as a multimodal assistant, meaning it works with more than one form of information. It understands and produces text, images, audio and data, and can respond with 3D models, interactive charts, animations, maps and small interactive programs.

Ant says the system uses collaborating agents, with different skills working in parallel to produce a single result for the user. An agent is a software component assigned to carry out a task. In this arrangement, several components can contribute to the same answer without requiring the user to manage them individually.

A study session illustrates how that could work. The assistant can extract the key points of a topic, arrange them in a logical outline and pair the explanation with 3D illustrations or short animations. For a data task, it can read a table and propose an interactive chart that changes when its inputs change.

These visual responses extend LingGuang beyond conventional text answers. A student can work with an explanation that includes diagrams or movement, while someone comparing figures can use a chart whose behavior reflects changes to the underlying information.

What the AGI camera is designed to do

LingGuang also includes a feature called AGI camera. AGI stands for artificial general intelligence. The camera feature lets the assistant analyze scenes through a phone camera and provide context or edits in real time.

A user could ask for the fastest way to measure the space in a living room, or request that the system remove the background from a video clip and brighten the subject. The same tool can generate images or short videos from text or an existing photograph.

Live camera processing brings privacy concerns alongside those capabilities. Images, and sometimes audio, may contain details about a person's home, surroundings or other people. Clear controls and safeguards on the device will matter as users decide what they are comfortable letting the assistant process.

Users will also need transparency about which information is processed on the phone and which is sent to cloud servers, with options to restrict storage and sharing. The description of the feature leaves those processing and storage arrangements unspecified.

A launch timeline shaped by demand

LingGuang's first days combined rising downloads, a brief service interruption and rapid movement through Apple's mainland China charts. The sequence shows how quickly interest in the new app turned into pressure on its infrastructure:

• Three days after launch, on Thursday evening, Flash App was temporarily suspended during peak demand while engineers expanded capacity. It returned the same evening.
• Within four days of launch, downloads passed one million, according to Ant Group.
• By Saturday, LingGuang ranked first among free utility apps and seventh among all free apps on Apple's mainland China App Store.
• By the sixth day, downloads had exceeded two million.
• Early the following week, the app rose to sixth among all free apps in the mainland China store.

Ant says its team expanded capacity in several waves as people tried creating and sharing their first apps. The immediate challenge was keeping the feature available while meeting demand for both initial app generation and repeated changes.

The company described users quickly moving from a first version into further edits. Some made more than 100 changes in a short window, while typical sessions involved about six rounds of conversation. That pattern helps explain why the service needed capacity for continuing work as well as new arrivals.

The team and models behind LingGuang

LingGuang was built by a team within Ant Group under the supervision of He Zhengyu, a PhD graduate of the Georgia Institute of Technology. The project has been in active development since March 2025.

Ant has also established an AGI Lab to develop its own large AI models. Other products it launched during 2025 include a healthcare assistant and a service robot, placing LingGuang within a wider set of efforts to bring AI into everyday services.

The app can work with several Chinese language models, including Alibaba Cloud's Qwen and selected open source models. Language models are the AI systems that interpret requests and generate responses. Open source models make their underlying resources available for others to use under their respective terms.

Ant executives describe LingGuang and Alibaba's Qwen app as complementary. Qwen focuses on a general chatbot experience at massive scale, while LingGuang concentrates on creating apps and visual outputs on mobile devices.

Cai Wei, LingGuang's technical lead and a former Google engineer, described the product's position in mobile software creation:

It is the industry's first mobile app that allows users to easily generate and customize apps.

China's AI competition moves into everyday tools

Alibaba's Qwen chatbot crossed about ten million downloads within a week of public testing, while ByteDance has pushed its assistants and media tools into leading app rankings. LingGuang's climb into the top tier of Apple's mainland China charts places it in a crowded competition for consumer attention.

Strong marketing helps these products attract initial interest. Their ability to keep users returning will depend on how often people find a reason to use them. For LingGuang, that means building or using small tools that solve recurring problems in daily life.

Parents are creating weekly planners with activity suggestions for children. Drivers are assembling calculators to compare fuel and maintenance costs between car models. Office teams are making quick training checklists for new employees, and students are experimenting with study guides that combine short text, diagrams and small quizzes.

Each example follows the same basic cycle: describe the desired tool, check what the assistant produces, then request changes. A family, class or team can receive the resulting app without its creator first having to learn the usual software development workflow.

The global growth of tools such as Replit shows that building software through natural language has an audience beyond China. LingGuang brings that approach into a mobile assistant, with a focus on immediately usable mini apps and visual responses.

Quality and privacy will shape everyday use

Automatically generated code can contain bugs or security gaps, especially when an app handles sensitive information. A tool may look polished yet fail when someone enters an unusual value or uses it in a situation its creator did not anticipate. Checking the result remains part of the process.

A calculator with an incorrect formula can produce a misleading answer, and a planner can give poor advice. AI responses can also contain mistakes or bias. Tasks involving health, finance or education will need strong evaluation because errors in those areas can affect consequential decisions.

Possible safeguards include vetted templates, human review of apps placed in public galleries and connections to external data that users explicitly choose to enable. Clear warnings, permissions and ways to upgrade an app will also matter as people move from experiments to tools they share and use regularly.

Camera features create another set of responsibilities. Users need understandable controls over the images and audio they allow the service to process, and over storage and sharing. China also has strict content and data rules for AI systems, so services growing at this pace must meet those requirements while continuing to introduce features.

LingGuang's first days show demand for an assistant that turns ideas into small, usable programs. Keeping those programs dependable, protecting the information they handle and maintaining enough server capacity are the practical tasks accompanying its rapid growth.

Key Points

• Ant Group says LingGuang passed one million downloads within four days and two million by its sixth day.
• Flash App creates a functioning mini app from a plain language request in about 30 seconds, then supports revisions through chat.
• A Thursday evening traffic spike briefly forced Flash App offline. It returned the same evening as Ant expanded capacity.
• LingGuang reached first place among free utilities and rose from seventh to sixth among all free apps on Apple's mainland China App Store.
• Typical sessions involve about six rounds of edits. Some early users made more than 100 changes in a short period.
• The assistant supports text, images, audio, data, 3D models, charts, animations, maps and camera based scene analysis.
• Development has been active since March 2025 under CTO He Zhengyu. The app works with models including Alibaba Cloud's Qwen and selected open source models.
• LingGuang is available through iOS, major Android stores and the web for users inside and outside China.
• Code quality, privacy controls, server capacity and compliance with China's content and data rules are central concerns as usage grows.