Mojo (language)

Definition

Mojo is a programming language developed by Modular that aims to be a superset of Python syntax, combining Python's usability with systems-programming performance features such as static typing, ownership-based memory management, and hardware-aware compilation, targeting AI and machine learning workloads that need both Python ergonomics and near-C performance.

Overview

Mojo was created by Modular, a company founded by Chris Lattner (creator of LLVM and Swift) and Tim Davis, with the stated goal of unifying AI research and production code under a single language. The motivating problem is a common pattern in machine learning: researchers prototype models in Python for its ergonomics and rich library ecosystem, but performance-critical parts (kernels, tensor operations) are then rewritten in C++ or CUDA for production speed, creating a two-language problem where research and production code diverge and require specialized engineers to bridge them. Mojo addresses this by aiming to be a true superset of Python syntax — existing Python code should, in principle, run as valid Mojo code — while adding opt-in systems-programming features layered on top: static type declarations for performance-critical code paths, an ownership and borrow-checking model influenced by Rust for memory safety without a garbage collector, and direct access to low-level hardware programming constructs for writing highly optimized kernels. Developers can start with ordinary, dynamically typed Python-style code and progressively add type annotations and low-level constructs only where performance matters, rather than rewriting an entire program in a different language. Under the hood, Mojo is built on MLIR (Multi-Level Intermediate Representation), the compiler infrastructure Lattner also helped develop, which allows Mojo to target diverse hardware backends including CPUs, GPUs, and specialized AI accelerators from a single source language, aiming to eliminate the need for separate CUDA or hardware-specific kernel code in many cases. Mojo is positioned specifically for AI/ML infrastructure, model inference, and numerical computing workloads rather than as a general-purpose language replacement for Python, and as of the mid-2020s remains under active development with an evolving specification, meaning it is not yet a drop-in, fully compatible Python superset and adoption is concentrated in performance engineering and ML infrastructure teams.

Key Features

• Designed as a syntactic superset of Python, aiming for Python code compatibility
• Opt-in static typing and ownership-based memory management for performance-critical code
• No mandatory garbage collector; Rust-influenced borrow checking for memory safety
• Built on MLIR compiler infrastructure to target CPUs, GPUs, and AI accelerators
• Aims to eliminate the 'two-language problem' of Python research code plus C++/CUDA production code
• Progressive typing: developers add performance features incrementally, not all at once
• Designed specifically for AI/ML numerical and kernel-level workloads
• Created by the team behind LLVM and Swift, emphasizing compiler-driven performance

Use Cases

Alternatives

History

Mojo was created by Modular, the company founded by Chris Lattner — the original architect of LLVM and the Swift language — together with Tim Davis. It was announced in May 2023, with the first publicly testable version made available through a hosted online playground rather than a local install. Mojo is designed as a superset of Python that keeps Python's familiar syntax and ecosystem while adding systems-programming features and the performance needed for AI and hardware-accelerated numeric workloads. Rather than targeting LLVM directly, Mojo is built on MLIR (Multi-Level Intermediate Representation), the compiler framework Lattner also helped create, which lets it target CPUs, GPUs, and specialized accelerators.

Frequently Asked Questions

Related in Programming