What is Modular?

Modular is a unified AI inference platform that spans from GPU kernels to cloud API endpoints. It includes MAX, a hardware-agnostic serving framework with an OpenAI-compatible API, and Mojo, a systems language for writing high-performance GPU kernels. The platform runs the same model and codebase across NVIDIA, AMD, Trainium, TPU, Qualcomm, Intel, ARM, and Apple silicon, and can be deployed in Modular's hosted cloud or in your own VPC.

How to use Modular?

Sign up or request a demo on modular.com to get started. Access frontier and open models through Shared Endpoints via an API, or deploy dedicated endpoints for mission-critical reliability. Run the MAX framework and Mojo self-hosted, or use Modular Cloud for fully managed, pay-by-usage serving. Use the Model Library and docs to port custom models and write custom GPU kernels.

Core features of Modular

Unified inference stack from GPU kernel to API endpoint
MAX serving framework with OpenAI-compatible API
Mojo systems language for high-performance GPU kernels
Hardware portability across NVIDIA, AMD, Trainium, TPU, Qualcomm, Intel, ARM, and Apple silicon
Shared and Dedicated Endpoints for frontier and custom models
Deployment in Modular Cloud or your own VPC
1000+ supported models including DeepSeek and Kimi
Agentic Skills and AI coding skills for model porting