AI底层基础设施的竞争，正在从模型训练转向编程语言层面的“核战争”。模块化（Modular）团队今天正式扔下一枚重磅炸弹：其新一代通用编程语言Mojo，连同编译器、工具链等核心组件，已全面开源，采用Apache 2.0协议（附带LLVM例外条款）。这标志着这门号称要“打破传统语言天花板、直插GPU与AI加速核心”的语言，终于向全球开发者敞开了大门。
Major progress has been made in the field of artificial intelligence’s underlying technologies. The Modular team has officially announced that its new-generation general-purpose programming language Mojo, along with its compiler, toolchain, and other core components, have been fully open-sourced after years of development. The project is released under the Apache 2.0 license with LLVM exceptions, marking a new stage in the development of this programming language, which aims to break traditional language limitations and fully enter the GPU and AI acceleration fields.
As an ambitious general-purpose programming language, Mojo integrates the latest achievements from the fields of compilers and programming language research, aiming to solve performance bottlenecks faced by traditional languages on modern hardware. Over the past four years, the language has been developed through an open community approach, but the compiler itself remained closed-source. With the recent release of the Mojo 1.0 version achieving source code stability, Modular has decided to open-source the entire compiler and toolchain today, allowing developers around the world to freely build and distribute related binary files.
In terms of open-source strategy, Modular has taken a cautious and gradual approach. The team leader stated that a small, tight design team is best suited for finding the “soul” of the language, while broad community feedback is key to breaking out of industry echo chambers. Therefore, the official has already open-sourced the Mojo standard library and gradually released tens of thousands of lines of kernel code, tools, and technical support. Now, with the compiler source code being launched on the official GitHub repository, developers can simply clone the project and use the Bazel build system to execute a single command to fully build the compiler or run tests.
To meet the actual needs of different developers, Modular offers flexible configuration options. Developers who do not need to modify the compiler source code can directly use pre-built nightly binary distributions, saving a lot of compilation waiting time. However, the official has clearly stated that for developers who need to customize MAX kernels or models, the pre-built compiler versions remain essential.
In terms of community contributions and future planning, the Mojo standard library has accepted community contributions since 2024. However, regarding contributions to the compiler and toolchain, considering the special management requirements of the AI programming era, the official has not yet opened direct submissions. It is expected that relevant contribution channels will be gradually opened by the end of this year, and technical questions encountered during reading the source code will be addressed through the official forum.
客观来看，Mojo开源的意义远不止于“又多了一个编程语言”。它直接挑战了Python在AI领域的统治地位——Python虽然生态丰富，但性能瓶颈让开发者不得不依赖C++/CUDA做底层优化。Mojo的目标是让开发者用Python般的语法写出接近C++的性能，同时原生支持GPU、TPU等异构硬件。但挑战同样巨大：社区贡献的门槛、编译器的成熟度、以及能否吸引足够多的第三方库开发者，都将是决定其成败的关键。目前来看，Mojo选择了一条“先闭源定型、再开源冲量”的稳妥路线，在AI底层工具链白热化的今天，这一步棋走得既大胆又谨慎。
