aid: parasail
specVersion: '0.1'
name: Parasail Rate Limits
url: https://docs.parasail.io/parasail-docs/
description: |
  Parasail enforces request-per-minute (RPM) ceilings by usage tier on its OpenAI-
  compatible inference endpoints. A token-volume limit is not currently enforced.
  Batch jobs are bounded by a 24-hour completion window rather than RPM. Dedicated
  deployments have no global RPM ceiling beyond what the provisioned replicas can
  serve.
tiers:
  - id: free
    name: Free
    rpm: 5
    description: Starter tier capped at 5 requests per minute across all inference endpoints.
  - id: user
    name: User
    rpm: 500
    description: Standard developer tier capped at 500 requests per minute.
  - id: dedicated-serverless
    name: Dedicated Serverless
    rpm: 1000
    description: Reserved-pool tier capped at 1,000 requests per minute.
  - id: dedicated-serverless-pro
    name: Dedicated Serverless Pro
    rpm: 4000
    description: High-throughput reserved-pool tier capped at 4,000 requests per minute.
  - id: enterprise
    name: Enterprise
    rpm: Unlimited
    description: Custom contracts; no platform-imposed RPM ceiling.
policies:
  - scope: inference
    endpoints:
      - https://api.parasail.io/v1/chat/completions
      - https://api.parasail.io/v1/completions
      - https://api.parasail.io/v1/embeddings
      - https://api.parasail.io/v1/models
    enforcement: RPM (per API key, per tier)
    tokenLimit: Not currently enforced
  - scope: batch
    endpoints:
      - https://api.parasail.io/v1/batches
    enforcement: 24-hour completion window per job
  - scope: dedicated
    endpoints:
      - https://api.parasail.io/api/v1/dedicated/deployments
    enforcement: Throughput bounded by provisioned replicas, not by global RPM