Modal · Rate Limits

Modal Labs Rate Limits

Modal governs usage through concurrency limits rather than classic requests-per-minute API throttling, because the developer surface is an SDK/gRPC control plane and user-deployed containers, not a metered REST API. The key limits are concurrent containers and concurrent GPUs, which scale by subscription tier. User-deployed web endpoints (*.modal.run) scale by container concurrency; any request-level rate limiting on those endpoints is defined by the developer's own function code.

Modal Labs Rate Limits is the machine-readable rate-limit profile for Modal on the APIs.io network, conforming to the API Commons Rate Limits specification.

It captures 6 rate-limit definitions, measuring containers, gpus, and requests.

The profile also includes 3 backoff/retry policies defined and response codes documented for throttled.

Tagged areas include Serverless, Compute, GPU, AI Infrastructure, and Rate Limiting.

6 Limits Throttle: 429
ServerlessComputeGPUAI InfrastructureRate LimitingQuotasConcurrency

Limits

Concurrent Containers (Starter) workspace
containers
100
Maximum simultaneously running containers on the Starter tier.
Concurrent GPUs (Starter) workspace
gpus
10
Maximum simultaneously attached GPUs on the Starter tier.
Concurrent Containers (Team) workspace
containers
1000
Maximum simultaneously running containers on the Team tier.
Concurrent GPUs (Team) workspace
gpus
50
Maximum simultaneously attached GPUs on the Team tier.
Concurrent Containers / GPUs (Enterprise) workspace
containers
negotiated
Higher GPU and container concurrency under an Enterprise agreement.
Web Endpoint Request Rate function
requests
user-defined
*.modal.run endpoints scale with container concurrency; any per-request rate limiting is implemented in the developer's own function code.

Policies

Autoscaling
Modal autoscales containers up to the workspace concurrency limit and scales to zero when idle; excess demand queues rather than being rejected.
Tiered Concurrency
Container and GPU concurrency ceilings raise from Starter to Team to Enterprise.
Backoff Strategy
SDK/CLI clients should implement exponential backoff with jitter and honor Retry-After on any 429 from the control plane.

Sources