specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Beam
providerId: beam-cloud
created: '2026-06-20'
modified: '2026-06-20'
reconciled: false
tags:
- Serverless
- GPU
- Python
- Inference
- Containers
- Rate Limiting
- Quotas
- Throttling
description: >-
  Beam governs workloads primarily through container concurrency limits rather than
  classic per-minute request quotas. Each usage tier caps the number of GPU and CPU
  containers that may run simultaneously (Developer 5 GPU / 30 CPU, Team 50 GPU /
  1,000 CPU, Growth custom / unlimited), and the platform autoscales deployments up
  to those ceilings. Synchronous web endpoints are additionally bound by an
  invocation time limit of roughly 180 seconds, beyond which work should move to
  asynchronous task queues.
notes: >-
  Concurrency ceilings come from the Beam pricing tiers; per-endpoint autoscaling,
  warm/keep-alive, and any platform-level request throttling should be verified in
  the Beam dashboard and docs during reconciliation.
sources:
- https://www.beam.cloud/pricing
- https://docs.beam.cloud/v2/scaling/concurrent-inputs
- https://docs.beam.cloud/v2/endpoint/overview
responseCodes:
  throttled: 429
limits:
- name: GPU Container Concurrency
  scope: account
  metric: containers
  limit: '5 (Developer) / 50 (Team) / custom (Growth)'
  notes: Maximum number of GPU containers running concurrently per tier.
- name: CPU Container Concurrency
  scope: account
  metric: containers
  limit: '30 (Developer) / 1000 (Team) / unlimited (Growth)'
  notes: Maximum number of CPU containers running concurrently per tier.
- name: Synchronous Endpoint Timeout
  scope: endpoint
  metric: seconds
  limit: ~180
  notes: Web endpoints target synchronous work under ~180 seconds; longer work belongs in task queues.
- name: Autoscaling
  scope: deployment
  metric: containers
  limit: up to tier concurrency ceiling
  notes: Deployments scale out per concurrent inputs up to the account's GPU/CPU concurrency limit.
policies:
- name: Concurrency-Based Throttling
  description: New invocations queue or scale containers up to the tier ceiling rather than being rejected outright.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter and honor Retry-After on 429 responses.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com