December 2025 Summaries

In 2025, Modular focused on creating a unified AI infrastructure capable of deploying AI across various hardware without compromises. Significant achievements included launching support for AMD MI355X, setting performance records with NVIDIA Blackwell, and powering Inworld TTS 1 as the leading speech model. The company open-sourced its Mojo standard library and MAX Python API, allowing developers insights into high-performance GPU kernels and model execution stacks. They launched Mammoth for scaling AI workloads, introduced educational resources for GPU programming, and released a new API for custom model creation. Community engagement thrived through hackathons and meetups, with contributions enhancing the Mojo ecosystem. Additionally, Modular secured $250 million to bolster its mission of making AI accessible and efficient across all infrastructures, setting the stage for continued growth into 2026.

The MAX Agentic Cookbook is a modern full-stack application designed to showcase the agentic AI capabilities of the Modular MAX as a comprehensive LLM serving solution. Built using FastAPI for the backend in Python and React with TypeScript for the frontend, the app is structured for developer familiarity and flexibility, featuring a modular architecture with distinct recipe components for functionalities like multi-turn chat and image captioning. It requires Python 3.11 or higher and Node.js 22.x or higher, with a setup that includes cloning the repository, configuring LLM endpoints, installing dependencies, and running the app using either VS Code or Docker. The backend leverages FastAPI and uvicorn, and includes various endpoints for health checks, listing recipes and models, and executing chat or image captioning tasks. The frontend utilizes React 18, Vite, and a range of technologies like React Router and Mantine for UI development, supporting features such as lazy loading and automatic data fetching. The app is designed to be run with a compatible GPU for optimal performance, and the repository includes guides for contributing, architecture, and Docker deployment, all under the Apache-2.0 WITH LLVM-exception license.

Mojo, a programming language developed by Modular, is nearing the release of its 1.0 version, marking a significant milestone in its development since its inception three years ago. Designed to target diverse hardware such as CPUs, GPUs, and other accelerators while using Python's intuitive syntax, Mojo aims to stabilize its ecosystem and support developers in building high-performance applications across various domains. The 1.0 release will focus on stabilizing core features and establishing a solid foundation with semantic versioning to ensure that current projects remain functional as new features are added. Although some advanced features like private members and a robust async programming model will be deferred to future updates, the development team plans to support incremental enhancements and backward-compatible features in subsequent 1.x releases. A future transition to Mojo 2.0 is anticipated, with efforts to ensure smooth compatibility between versions. The Modular team invites community participation in the open development process, encouraging feedback and collaboration to refine the language further.

The Modular universe is experiencing significant growth and innovation, with community members around the globe developing projects that enhance AI and performance capabilities using MAX and Mojo. Notable contributions include a highly efficient FFT implementation and a new qubit simulator, both demonstrating Mojo's superior performance. Modular's impact extends to AI and high-performance computing, with benchmarks showing Mojo's efficiency exceeding that of CUDA and other DSLs, and the Modular Platform powering top-tier performance in text-to-speech benchmarks. Key figures like Chris Lattner and Tim Davis have highlighted the platform's role in advancing scalable AI systems and bridging CPU and GPU compute, while Modular continues to foster an active community through events, open-source contributions, and educational content. The release of Modular Platform 25.7 introduces enhancements such as a fully open MAX Python API and improved Mojo GPU programming, reflecting the platform's commitment to reducing infrastructure burdens and propelling AI innovation.