Today, large-scale domestically developed models such as Kimi and DeepSeek are entering real-world development scenarios. Writing functions, fixing bugs, reading documentation, generating scaffolding, and even participating in longer development tasks are no longer just capabilities shown in demos.
But as developers know, writing code does not mean the project is complete.
A demo can be generated quickly, but a real project has to deal with many more details: whether the dependencies can run, whether edge cases have been handled, whether it can be maintained later, and whether the code can be reused by others.
This is why the more AI programming moves into real-world scenarios, the more it needs languages, tools, and ecosystems to support it.
The development of MoonBit over the past year provides a good example. It is not only improving a domestic programming language but also demonstrates how a new language can move from “being able to write” to “being able to use” through practices such as package ecosystems, AI-generated C compilers, AI programming evaluation, external platform access, browser experiments, developer tools, and formal verification.
MoonBit’s Development Over the Past Year
Looking back over the past year, MoonBit’s development path is quite clear: first, get developers to actually use it, then move into more complex engineering scenarios, and finally add the trusted code component.
In June 2025, MoonBit entered the Beta phase. This means it began to move from “something to try out” to “something to use seriously”.
From an ecosystem perspective, MoonBit’s official package ecosystem, mooncakes.io, already boasts nearly 9,000 packages, approaching the number of packages listed in Apple’s Swift Package Index, with cumulative downloads reaching 3.5 million. For a rapidly growing new language…
This is an important signal that a language is moving from a “project” to an “ecosystem.”
Complex basic software scenarios
In 2026, the MoonBit team showcased the Fastcc case study. Fastcc is a C compiler project built from scratch using an AI agent based on MoonBit. It doesn’t simply generate a function, a webpage, or a demo; instead, it involves AI in the development of a complete compiler project: from source code parsing and semantic processing to code generation, debugging, and performance optimization, the entire process is integrated into the engineering workflow. The project produced approximately 35,000 of lines structured code within 10 days and achieved self-hosting.
What’s truly noteworthy about this is that it elevates AI programming from “generating code snippets” to “participating in complex software engineering.” Compilers themselves are typical foundational software projects, demanding high levels of language expressiveness, code organization, and engineering tools. The emergence of Fastcc proves that MoonBit already possesses the capability to undertake complex engineering experiments.
AI programming evaluation
This direction has also extended to AI programming evaluation. The SWE-AGI benchmark, released in February 2026, chose MoonBit as its task language, requiring AI agents to implement engineering modules such as parser, interpreter, binary decoder, and SAT solver according to specific specifications. Each task involves 1,000 to 10,000 lines of core logic. Its significance lies in shifting the assessment of AI programming ability from “whether it can write an answer” to “whether it can complete a maintainable software project.”
External platforms are also beginning to incorporate MoonBit into their engineering systems. Golem Cloud, a US-based cloud service provider, is a cloud platform for WebAssembly Components, focusing on durable computing and agent runtime scenarios. In version 1.5, Golem Cloud provides first-class MoonBit support, enabling the writing of durable agents using MoonBit and their compilation into WebAssembly components that run on Golem. Its documentation covers MoonBit use cases such as HTTP endpoints, authentication, OpenTelemetry, LLM/AI capabilities, database access, webhooks, and scheduled tasks, and the related SDK has been released on mooncakes.io.
System Software Experiment
Overseas developer mizchi has long focused on front-end tools, browser infrastructure, and developer tools. Recently, he’s been working on a custom TUI browser project using MoonBit, attempting to bring webpage reading capabilities to the terminal environment. The previously released v0.0.2 version could already display basically readable content and reduced display misalignment; in the latest update, this custom TUI browser can now read HackerNews in the terminal, and the interactive experience is still being refined.
At a deeper level, mizchi also developed Crater based on MoonBit. This is a CSS layout engine written in Pure MoonBit, aiming not to create a complete browser, but to delve into a very core part of the browser rendering chain: layout calculation. For a browser to transform HTML and CSS into a page on the screen, it must first calculate the correct position and space occupied by each element. Crater provides precisely this fundamental capability.
MoonBit’s Exploration
MoonBit’s exploration in the Web field has gone beyond simply “compiling to Wasm” or “running a front-end example”; it has touched upon more fundamental system software experiments such as browser layout, terminal rendering, and lightweight browsers.
There are also new attempts in the area of developer tools. In January 2026, MoonBit demonstrated its static analysis practices. Static analysis can check for potential problems before the program runs, and is commonly found in compilers, IDEs, and code quality tools; MCIL can be understood as an intermediate layer used by MoonBit to provide this type of analysis capability.
On the application side, MoonBit also has examples like the CRDT real-time collaborative editor. CRDT is an algorithm commonly used for multi-user collaboration and offline synchronization. For example, when multiple people are editing documents simultaneously or merging content after a network outage, it is necessary to handle state consistency and conflict merging issues.
This case illustrates that MoonBit is not only suitable for system software experiments, but can also be used to express complex application logic.
Community contribution basic tools
The well-known Chinese open-source developer justjavac has open-sourced moonbit-case, which provides string naming style conversion capabilities, such as snake_case, camelCase, and kebab-case. It’s not a large-scale system project, but these kinds of small tools are important to the language ecosystem because real-world development relies heavily on these high-frequency, fragmented, and reusable foundational libraries.
Entering 2026, MoonBit’s path continues to extend towards stability and trusted code. The 0.8.0 release in February is officially regarded as an important milestone towards stability and production use; in April, version 0.9 further formal elevated verification to an important capability.
With AI-generated code becoming increasingly prevalent, the real scarcity lies not just in the code itself, but in the ability to verify that critical logic is sound. MoonBit’s introduction of formal verification capabilities into its language and toolchain is precisely a response to this issue.
From C compilers to AI programming evaluation, from external platform access to the TUI browser, from browser layout engines to static analysis, CRDT collaborative applications, and formal verification, MoonBit’s keywords over the past year have no longer been simple version iterations, but rather the advancement of ecosystem growth, the implementation of complex engineering projects, and the development of trustworthy code capabilities.
Starting with real projects, let’s build a domestic basic software ecosystem together
The China Computer Federation (CCF) (8th CCF Open Source Innovation Competition) and IDEA Research are jointly hosting the 2026 MoonBit Domestic Basic Software Open Source Competition. The competition focuses on building the MoonBit ecosystem and encourages developers to create open-source projects with engineering value in areas such as tool libraries, Wasm components, AI agents, compilers, browsers, and data processing. The total prize money is 150,000 RMB. Outstanding projects that reach the finals and win awards will receive certificates of honor from IDEA Research and the CCF.
If you are also interested in domestically developed basic software, next-generation programming languages, and AI programming practices, you are welcome to register for the competition and work together to add a truly usable piece to the MoonBit ecosystem.
Related links:
- MoonBit, a website featuring the eco-friendly mooncakes.io
https://mooncakes.io/
- MoonBit 0.9 published an article: Formal Verification Capabilities
https://www.moonbitlang.com/blog/moonbit-0-9-release
- AI-driven development of the C compiler Fastcc
https://www.moonbitlang.com/blog/fastcc-ai-driven-development
- SWE-AGI: A Software Engineering Benchmark for AI Agents
https://arxiv.org/abs/2602.09447
- Golem Cloud MoonBit Documentation
https://learn.golem.cloud/how-to-guides/moonbit/golem-add-moonbit-package
- mizchi’s custom TUI browser v0.0.2
https://x.com/mizchi/status/2010031878586413197
