Modal
Serverless cloud platform for running AI inference, training, sandboxes and batch jobs on GPUs.
Modal is a serverless cloud platform for AI workloads. Modal is used by writing Python code that the platform runs in autoscaling containers on CPUs and GPUs, covering model inference, fine-tuning, sandboxes for agents and batch jobs. Resources scale to zero when idle and are billed by the second.
Best for Engineering teams that want GPU compute for inference, training or agent sandboxes on demand, without provisioning or managing their own clusters.
- Listed in
- DevOps & MonitoringDeveloper API
- Platforms
- Web, API
- Pricing model
- Freemium
- Free plan
- Yes
- API
- Yes
- Made by
- Modal
- Added
Last verified: . Information taken from modal.com.
What is Modal?
Modal is a serverless cloud platform for teams that build, train and serve AI applications. Developers write Python, and Modal runs the code in containers that start within seconds and scale down to zero when idle. The platform is organised around a few products: Inference for serving language, image, video and audio models; Training for fine-tuning and reinforcement learning on single or multi-node GPUs; Sandboxes for running coding agents and untrusted code in isolation; and Functions, Notebooks and batch jobs for general compute. Underneath sit a custom container runtime, a distributed filesystem for model weights and datasets, and a capacity pool drawn from more than twenty clouds across many regions. Usage is metered per second for GPU, CPU and memory. Enterprise features include role-based access, audit logs, data residency controls, SOC 2 Type II compliance and HIPAA readiness.
Key Features
Model Inference
Serves LLMs and media models through endpoints, including OpenAI-compatible ones, for real-time or batch traffic.
Sandboxes
Creates isolated environments for agents and untrusted code, with snapshot and restore for quick starts.
Training and Fine-Tuning
Runs LoRA, supervised and full fine-tunes or RL jobs on single or multi-node GPU capacity.
Autoscaling Functions
Fans Python functions out across many containers and exposes them as HTTPS endpoints that idle at zero.
Storage for Models
Provides a distributed filesystem that loads images, weights and datasets wherever functions run.
Observability
Shows logs and metrics per function and container, with export to OpenTelemetry providers.
Screenshots
Use Cases
- Serving an Open LLM
- A team deploys an open model behind an endpoint that scales with request volume.
- Running Coding Agents
- An agent product executes generated code inside sandboxes instead of on its own servers.
- Batch Transcription
- A data team processes a backlog of audio files in parallel across GPUs.
- Image and Video Generation
- An app backend calls media models hosted on Modal for each user request.
- Research Fine-Tuning
- Researchers launch fine-tuning or RL runs without reserving a cluster in advance.
Things to know
- Idle time can be billed
- Containers stay alive for 60 seconds after the last input by default, and that window is charged unless you shorten it.
- Minimum container size
- Every container requests at least 0.125 CPU cores and 128 MiB of memory, and the charge follows whichever is higher, the requested or the used amount.
- Starter workspaces hold three people
- The Starter plan allows up to three seats; the Team plan removes the seat limit.
- Credit programmes exist
- Startups and academic researchers can apply for free compute credits.
FAQs
What is Modal?
Modal is a serverless cloud platform that runs AI inference, training, sandboxes and other compute-heavy Python workloads on autoscaling CPU and GPU containers.
What is Modal used for?
Modal is used to serve models, fine-tune them, run coding agents in sandboxes and process batch jobs without managing servers or clusters.
Is Modal free?
Modal can be used without a base fee on its Starter plan, which comes with a monthly allowance of free compute credits; usage beyond that is pay-as-you-go.
How much does Modal cost?
Modal charges per second for the GPU, CPU and memory you use, on top of Starter, Team or Enterprise plans. See the official pricing page for current rates.
How does Modal bill usage?
Modal bills while an application loads and while containers process inputs, plus a configurable idle window that defaults to 60 seconds. Nothing is charged once an app scales to zero containers.
What can I deploy on Modal?
Modal says it can run any containerizable application written with Python, for example ML inference, training, data pipelines and job queues.
Is Modal suitable for regulated workloads?
Modal reports SOC 2 Type II compliance and describes itself as ready for HIPAA workloads, with audit logs, RBAC and data residency controls.
Can I use AWS or GCP credits on Modal?
Not directly. Modal says cloud credits cannot be applied, though committed spend can be used through the AWS and GCP marketplaces.
Who it's for
- ML engineers
- AI startups
- Platform teams
- Researchers
- Agent developers
Comments
No comments yet. If you have used Modal, tell other readers how it went.
Alternatives to Modal
| Tool | Free plan | Starting price | API |
|---|---|---|---|
 LangChainOpen source agent frameworks plus the LangSmith platform for tracing, evaluating and deploying AI agents. |
Yes | — | Yes |
 VercelCloud platform for deploying web apps and AI agents, with an AI gateway, sandboxes, workflows and a global CDN. |
Yes | $20/mo | Yes |
 AtomDomain marketplace and registrar with AI tools for generating, testing and contesting business names. |
— | — | Yes |
 BannerbearAPI that generates images, PDFs, animations and videos automatically from reusable design templates. |
— | $49/mo | Yes |
 BigjpgAI image enlarger that upscales artwork and photos up to 16x while reducing noise and jagged edges. |
Yes | $6 | Yes |
Listed in the same categories as Modal. A starting price appears only where we checked it on the maker’s pricing page; a dash means not checked or not published. All DevOps & Monitoring tools