Modal

Running the code around a model

Modal provides a cloud execution platform for developers working in Python. Its documented uses include GPU inference, parallel batch jobs, model training and fine-tuning, and isolated sandboxes for executing generated code. Teams can bring open weights or custom models rather than being limited to a catalog of hosted model APIs. [1]

That makes Modal part of the infrastructure between a model and an application: it runs the functions and compute that a team supplies. Inference serves model outputs; batch jobs process many inputs; sandboxes execute code in isolated environments. Those are separate workloads supported by the same platform, and do not imply that Modal authored the models being run. [1]

Opening access beyond a waitlist

On October 10, 2023, Modal made its platform generally available, opening immediate registration after inviting users from a waitlist. At launch it supported autoscaling GPU endpoints, batch execution, scheduled jobs and web endpoints, with images, hardware and persistent storage specified in Python. Contemporary coverage corroborates the public launch date. [2][3]