Modular Open-Sources Mojo 1.0 and MAX Source, Completes Acquisition by Qualcomm

• Modular open-sourced the systems language Mojo 1.0 under the Apache 2.0 license and released the source code for the inference framework MAX
• Mojo 1.0 supports NVIDIA, AMD, and Apple hardware without CUDA dependencies, and has released its compiler and standard library
• The MAX framework released its graph compiler and runtime code but adopted a Source-Available model by applying the MAX License, which restricts the disclosure of modified sources
• On the NVIDIA B200, BF16 matrix multiplication performance reached 1772.9 TFLOPS, which is 100.6% of cuBLAS, and the first token time for Gemma 4 31B was reduced to approximately 30 seconds in immediate execution mode

Technical Features and Licensing

• Mojo supports static typing and AOT compilation, ensuring memory safety without a GC through Ownership-based memory management
• MAX provides an OpenAI-compatible inference server and natively supports continuous batching and KV cache management
• Due to the MAX License applied to MAX core components, commercial use is permitted, but disclosure of modified sources is limited to forks for contribution purposes only
• Official Windows support is currently limited to WSL, with native support announced through collaboration with Microsoft