Modal
6.7/10Modal is a serverless cloud platform that runs Python code on GPUs and CPUs with per-second billing, aimed at teams building and serving AI applications.
Data last updated November 2026 · Our methodology
Overview
Modal is a serverless compute platform for AI and data workloads. Developers write Python functions with decorators, and Modal builds the container image, schedules it on CPU or GPU workers, and scales from zero to thousands of containers. Functions can run as one-off jobs, scheduled cron jobs, batch maps, or deployed web endpoints. Sandboxes for running untrusted or agent-generated code are a growing part of the product. GPUs range from the T4 to the B200 and B300, and no quota requests or reservations are needed. It suits ML engineers, AI startups, and product teams that run bursty inference, fine-tuning, or batch processing and do not want to manage Kubernetes or GPU clusters. The pricing has two parts. A plan fee ($0 for Starter, $250 a month for Team, custom for Enterprise) comes with a monthly compute credit of $30 or $100. Usage is then billed per second for GPU, CPU core, and memory, with extra charges for volumes, egress, and per-token Shared Endpoints. Pinning a region multiplies base prices by 1.15 to 1.75, and non-preemptible execution costs 3 times base. There are Python, JavaScript/TypeScript, and Go SDKs, a CLI, OpenTelemetry trace export, custom domains, and AWS and GCP marketplace purchasing for Enterprise. The main limitation is cost for steady workloads. Always-on services, keep-warm containers, and multi-day training runs can cost more than reserved GPU rentals. Idle container time after a request is also billed, so bills need careful modelling.
Our Rating
Hosting & Infrastructure6.8/10
AI Software6.7/10
Who Modal is best for
Good fit
It is built around a Python SDK and CLI, so infrastructure is defined in code.
Scale to zero, monthly credits, and a startup credit program fit early AI companies.
The Enterprise plan adds SSO, audit logs, HIPAA support, volume discounts, and marketplace purchasing.
It works well for SaaS products that serve AI inference endpoints that scale with demand.
Less suited
Academic credits of up to $10k support researchers and labs.
Pros & Cons
Pros
- GPUs from the T4 up to the B300 are available without quota requests or reservations, and workloads can burst to thousands of containers.
- Per-second billing and scale-to-zero mean idle bursty inference workloads cost nothing.
- Python developers define infrastructure in code and deploy with a single CLI command, with no Kubernetes or Dockerfiles needed.
- The free Starter plan includes $30 a month of compute, and startup and academic credit programs are available.
- Sandboxes support large numbers of concurrent isolated environments for AI agents and untrusted code.
Cons
- Always-on or steady workloads can cost more than reserved GPU rentals from providers such as RunPod or Lambda.
- Containers stay billable for 60 seconds after the last request by default, and keep-warm settings are a common source of bill surprises.
- Region pinning raises prices by 1.15 to 1.75 times, and non-preemptible execution by 3 times.
- SSO, audit logs, and HIPAA support are limited to the custom-priced Enterprise plan.
- Credits from AWS, GCP, or Azure cannot be applied to Modal usage.
Key Features
Not included
- Managed WordPress
- Dedicated Servers
- Shared Hosting
Modal Pricing
Starter
- Up to 3 seats
- 1-day log retention
- 1 TiB egress included
- Community Slack support
- $30/month compute credits
- 200 deployed apps, 5 crons
- 100 containers, 10 GPU concurrency
Team
- Unlimited seats
- 30-day log retention
- 10 TiB egress included
- $100/month compute credits
- 1,000 deployed apps, unlimited crons
- 5,000 containers, 50 GPU concurrency
- Custom domains, static IP proxy, RBAC
- Rollbacks to 3 versions, environment budgets
Enterprise
- Volume discounts
- 100 TiB egress included
- Okta SSO, audit logs, HIPAA
- AWS and GCP marketplace committed spend
- Custom credits and higher GPU concurrency
- Private Slack and embedded ML engineering support
Frequently Asked Questions
Yes. The Starter plan has no platform fee, includes $30 of compute credit per month, and allows up to 3 seats.
GPU, CPU core, and memory time are billed per second. Volumes, network egress, and per-token Shared Endpoints are billed separately, on top of the plan fee.
Yes. You can pin broad regions such as the US, EU, or Asia-Pacific, or narrower regions and countries. Broad regions cost 1.15 times base prices and narrow regions 1.75 times.
SOC 2 compliance is listed for all plans. HIPAA support, SSO, and audit logs require the Enterprise plan.
Starter and Team users get the community Slack. Enterprise customers get a private Slack channel and embedded ML engineering support.
Enterprise customers can use committed spend through the AWS and GCP marketplaces. Existing AWS, GCP, or Azure credits cannot be applied to Modal usage.
Similar Tools
See all alternatives
Amazon's public cloud platform, with over 200 pay-as-you-go services for compute, storage, databases, networking and AI, used by startups through large enterprises and governments.
Google's public cloud platform providing pay-as-you-go virtual machines, Kubernetes, serverless, databases, storage, networking, data analytics and AI services for businesses of every size.
Northflank is a Kubernetes-based developer platform for deploying services, jobs, databases and GPU workloads on its managed cloud or on customers' own cloud accounts.