Modal

What it does

Modal takes Python code (packaged into a container) and runs it on cloud GPUs and CPUs, scaling containers up with traffic and back to zero when idle. Inputs are application code, container images, model weights and datasets; outputs are running functions, HTTPS endpoints, batch job results and logs and metrics. It also offers Sandboxes, isolated environments for running untrusted or agent-generated code, and OpenAI-compatible inference endpoints.

Developers use it to deploy ML inference, fine-tune or train models, run compute-heavy batch jobs, and host agent sandboxes. Its distinct features are per-second billing for GPU and CPU time, autoscaling across capacity pooled from multiple clouds, and a distributed filesystem for loading large models quickly. Setup requires a Modal account, the Python package and an authentication step.

Modal suits developers and ML teams who want to avoid managing Kubernetes or cloud infrastructure directly. A free Starter plan includes $30 of monthly compute and up to three seats, with Team and Enterprise plans for larger organizations. Limitations: Python is the primary language, Shared Endpoints tokens are billed from the first request, and enterprise pricing requires contacting Modal.

What it can do

• Run Python functions on cloud GPUs and CPUs with autoscaling to zero
• Create isolated Sandboxes for running untrusted or agent-generated code
• Turn functions into HTTPS API endpoints that scale with traffic
• Run large-scale batch jobs and fan out work across many GPUs
• Serve OpenAI-compatible inference endpoints and custom inference stacks
• Built-in logs and metrics per function and container, with OpenTelemetry export

Works with

• OpenTelemetry (export to any OpenTelemetry provider)
• AWS and GCP cloud marketplaces (committed spend)
• Okta SSO (Enterprise)

Example workflows

Deploy a GPU inference function

Illustrative example. Input: a Python function and a container image with a transformers library. Steps: decorate the function with a GPU type and image, then run modal run on the file. Output: the function executes on a cloud GPU and returns generated text.

Run untrusted code in a Sandbox

Illustrative example. Input: code produced by an agent or user. Steps: create a Sandbox from a Modal image, execute the code inside it, then terminate it. Output: the code result, with the Sandbox isolated from the host environment.

Before you choose

• Loading and idle-timeout periods can incur compute charges.
• Shared Endpoint token usage is separate from monthly compute credits.
• Region multipliers, resource requests and Sandbox CPU/memory pricing affect the final bill.

Listed under

• Coding
• Sandboxes and infrastructure
• AI inference
• AI infrastructure
• Developers
• Fine-tuning
• GPU cloud
• GPU compute

Get started

Sign up at modal.com, install the modal Python package, then run modal setup to authenticate.

1. Create an account at modal.com.
2. Run pip install modal to install the Python package.
3. Run modal setup to authenticate (or python -m modal setup if that fails).

Starter has no platform fee and includes $30/month of compute credits for up to three seats. Team costs $250/month plus compute, with $100/month of credits. Shared Endpoints are billed per token from the first request and are not covered by compute credits. Compute includes loading and the configured idle timeout. Sandbox CPU and memory have a 3× non-preemptible rate; GPU rates are standard.

Source: modal.com/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Modal: The platform for production AI ↗modal.com
• Pricing: Pay per second for GPUs and CPUs | Modal ↗modal.com
• Introduction | Modal Docs ↗modal.com

Company website

modal.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.