Skip to content

Cloud & Architecture · Freemium

Modal

Modal is a serverless cloud platform for running Python code on GPUs and CPUs without managing servers or Kubernetes. Developers define functions and container images in Python, decorate them with the resources they need, such as a GPU type, and deploy them as web endpoints, scheduled jobs or batch workloads. Modal positions the platform for AI inference, training and fine-tuning, batch processing and isolated sandboxes for coding agents.

Who it is for: teams that want to serve open-weight or custom models, run evaluation and batch jobs, or give AI agents sandboxed execution, and that would rather write Python than manage GPU instances, autoscalers and container registries.

How it works: Modal builds the container image, schedules containers on its own multi-cloud capacity pool and scales the number of containers with request volume, down to zero when there is no traffic. Its documentation says containers boot in about one second, while total cold start time also depends on how long your code takes to load model weights. Settings such as scaledown_window (how long an idle container stays alive, 60 seconds by default, configurable from 2 seconds to 20 minutes), min_containers and buffer_containers trade idle cost for lower latency. Memory snapshots can shorten repeated initialization work.

Pricing type: usage-based, billed per second, with a free monthly credit. As of November 11, 2026, Modal's pricing page lists GPUs from the Nvidia T4 to the B300, including H100 at $0.001097 per second (about $3.95 per hour) and A100 80 GB at $0.000694 per second (about $2.50 per hour), with CPU cores and memory billed separately. The Starter plan has no monthly fee and includes $30 of compute a month and 10 concurrent GPUs; the Team plan costs $250 a month with $100 of included compute and 50 concurrent GPUs; Enterprise is custom. Modal bills container load time and the idle window after the last request. Functions are preemptible by default, and the pricing page lists a 3x multiplier for non-preemptible execution and 1.15x to 1.75x for region selection. Modal also sells per-token Model APIs for hosted open-weight models.

Alternatives: Runpod Serverless, Replicate, Baseten, Google Cloud Run with GPUs, and managed model APIs such as Cloudflare Workers AI.

Official site: https://modal.com

Official site