Provider pricing

Modular MAX AI API pricing

Modular MAX is a unified inference platform — runtime, compiler, and serving — that runs PyTorch, ONNX, and Mojo models across NVIDIA, AMD, and CPU hardware. Self-hosted (free) or MAX Cloud with per-token billing on Llama, Mistral, Qwen, and custom checkpoints.

Founded2022
HeadquartersLos Altos, CA
Tracked modelsDirectory

Best for

Infra teams that want vendor-neutral, high-throughput inference for open models.

Strengths

  • 2-5x faster inference than vLLM on the same hardware
  • Hardware-agnostic: NVIDIA, AMD, Intel, Apple Silicon, CPU
  • Mojo language for custom kernels without writing CUDA
  • Free self-hosted tier; pay only for MAX Cloud usage