The platform for production AI
Cloud infrastructure for teams that develop, train, and serve AI applications at scale.
Trusted by the world’s leading AI teams
Container infrastructure that won’t box you in
Inference
Accelerating inference from endpoint to engine
Build, serve, and scale inference apps however your workload demands.
Access open intelligence
OpenAI-compatible endpoints built for out-of-the-box performance.
Deploy optimized models
Tune endpoints for the shape of your traffic.
Serve custom stacks
Bring your entire inference stack. Modal handles the infrastructure.
Training
Reinforcement learning and research at scale
From single-GPU fine-tuning to multi-node runs, Modal handles capacity and cluster management.
Fine-tune on any framework
RL, SFT, LoRA, full fine-tunes with instant access to the resources you need in the shape you need them.
Battle-tested OSS infra
Designed to work out of the box with open source frameworks like slime and Miles.
Elastic single or multi-node
Autoscaling, nearly instant access to single and multi-node GPU capacity.
Sandboxes
Computers for agents
From long running RL rollouts to interactive coding agents, Modal Sandboxes are secure, flexible, and built to scale.
Built for concurrency
Programmatically create millions of concurrent sandboxes in under a minute for fast rollouts.
Fast on any image
Bring any custom image or custom dependencies. Snapshot and restore for fast startup.
Adaptive compute resources
Burst CPU and memory resources to meet demand without paying overages or degrading service.
Functions
Build complex, compute-heavy applications.
Cloud compute that feels like local development. Scale any arbitrary code, anytime you need it.
Massively parallel compute
Fan out over thousands of GPUs, fully parallel, no job orchestration to manage.
Async batch jobs
Spawn async batch processing and long-running work in the background.
API gateways
Turn any function into an HTTPS endpoint that scales with traffic and idles at zero.
Infrastructure reimagined for the demands of AI workloads
Powerful primitives for end-to-end systems
Functions for serverless compute, Sandboxes for running untrusted code, and Servers for low-latency LLM inference, plus tools for state and networking.
CDN-style storage for models, weights, and more
A distributed filesystem for loading container images, model weights, and datasets at full bandwidth, everywhere your Functions run.
Fast container scheduling and startup
Modal runs a custom container runtime and image system built for large images and GPU scheduling.
Capacity that scales with your ambition
Globally distributed capacity
Infrastructure spanning 20+ clouds. Modal routes your workload to the right hardware, in the right region, automatically.
True, on-demand autoscaling
Burst to thousands of GPU and CPU resources at peak traffic, then back to zero when you're done. Only pay for what you use.
Any GPU, any time
Access the right resource for your workload in seconds without quota requests or long lead times.
Automated fleet health
Every node runs continuous passive and active health checks. Bad hardware is drained and replaced before it touches your workload.
Production-ready from day one
Everything you need to build and run AI workloads in production, from writing code to accessing compute, scaling across thousands of GPUs, and monitoring everything along the way.
Security and compliance for enterprise teams
Workload isolation by design, configurable network and endpoint access, granular RBAC, audit logs, and data residency controls. SOC 2 Type II compliant and HIPAA ready.
Out-of-the-box observability
Comprehensive logs and metrics for every function and sandbox, down to the container. For custom traces or tools like Datadog, export to any OpenTelemetry provider.
- Time
- 05:17am
- Containers
- 108
- GPU Utilization
- 94%
Maximum GPU utilization, billed by the second
GPUs when you need them, without quotas or reservations. You pay only while your code runs, with granular cost and usage controls built in.
Customers
“At DoorDash, we have to ship quickly and securely. Our AI platform supports teams across the company building internal tools and customer-facing apps, with inference, sandboxes for agents, and compute for custom code. Modal is a big part of how we get those apps from experiment to production.”
Latest updates
Modal Clusters are generally available
Multi-node GPU clusters with RDMA, gang scheduled from Modal's shared capacity pool and billed by the second, behind a single decorator.
Quail: Speeding up AI-SQL by jointly optimizing query planner and inference engine
Maximizing perf on AI-SQL queries with the KV-optimal left-deep join
How to serve trillions of tokens for trillion-parameter coding agents
Learn how we optimized performance and efficiency serving the workload that is changing software engineering forever — and you can too.