MAX
Open-source AI serving and modelling framework: a Python API, model pipelines and GPU kernels for NVIDIA, AMD and Apple hardware.
What it does
- Model inference
- GPU programming
- Model hosting
Features 4
- ServingServes hundreds of models out of the box through OpenAI-compatible inference APIs.
- ModelingProvides PyTorch-like model APIs and AI coding skills for porting custom models.
- GPU programmingOffers one API library and hundreds of open-source kernels covering CPUs, GPUs and ASICs.
- Benchmarking CLIIncludes a benchmarking CLI adapted from vLLM that runs against public or custom datasets.
Sources
- Pricing
-
Enterprise tier named on Modular's own pricing page as a custom engagement covering MAX and Mojo; the open-source claim is quoted from max.modular.com.
- Description
-
Official documentation · 18 Sep 2026
The page's own heading: 'MAX: A high-performance AI serving and modeling framework.'
- Constraints
-
Derived from the security facts on this page, from the same source.
- Security
-
Trust center read by curl 2026-10-22: 'Modular's SOC2 Type 2 report for 2026' under the unified inference platform (MAX and Cloud); Mojo is a language that runs on the customer's side and is not written. Subprocessors BentoML control plane (United States), HydraHost; no model provider.
- Sold within
-
Official documentation · 18 Sep 2026
Installed and run on its own from the package managers or the GitHub repository.
- URL
-
Official documentation · 18 Sep 2026
Fetched 2026-09-18, 200; modular.com/max 301-redirects here.
- Interfaces
-
Official documentation · 12 Oct 2026
Page: 'Specify the model with the CLI or docker container arguments. / Use standard OpenAI-compatible API calls from any client'.
- Intended for
-
Official documentation · 13 Oct 2026
Page: 'build your own with the MAX Python API and write custom kernels in Mojo' / 'Schedule a demo of MAX for enterprise to explore custom deployments built around your models and performance goals'.
Also from Modular
-
Mojo Developer tool
Apache-2.0 systems language for writing fast code across CPUs, GPUs and other accelerators without vendor lock-in, published by Modular.
-
Modular Cloud Infrastructure service
Modular's hosted inference service — shared and dedicated frontier-model endpoints, and the same stack deployed into a customer's own VPC.
Something wrong here? Send a correction — quote this product id: modular-max.