generated: '2026-08-14'
method: searched
source: https://docs.parallel.ai/getting-started/rate-limits
docs:
- https://docs.parallel.ai/getting-started/rate-limits
- https://parallel.ai/pricing
scope: per-account (API key)
counting_rule: >-
  Rate limits apply only to POST requests that CREATE new resources. GET requests that
  retrieve results or poll status do not count against the quota - e.g. polling
  GET /v1/tasks/runs/{run_id} does not consume the Tasks limit.
limit_count: 7
limits:
- product: Search
  operation: POST /v1/search
  limit: 600
  window: 1m
  scope: per-account
- product: Extract
  operation: POST /v1/extract
  limit: 600
  window: 1m
  scope: per-account
- product: Tasks / TaskGroups
  operation: POST /v1/tasks/runs or POST /v1/tasks/groups/{taskgroup_id}/runs
  limit: 2000
  window: 1m
  scope: per-account
- product: Chat
  operation: POST /v1beta/chat/completions
  limit: 300
  window: 1m
  scope: per-account
- product: FindAll
  operation: POST /v1beta/findall/runs
  limit: 300
  window: 1h
  scope: per-account
  note: >-
    The docs rate-limit table states 300 per hour for creating a FindAll generator; the
    marketing pricing page shows 25/hr for Find All. Both are published by Parallel and
    they disagree - recorded as found, not reconciled.
- product: Entity Search
  operation: POST /v1beta/findall/entity-search
  limit: 600
  window: 1m
  scope: per-account
- product: Monitor
  operation: POST /v1alpha/monitors
  limit: 300
  window: 1m
  scope: per-account
- product: Responses
  operation: POST /v1/responses
  limit: 300
  window: 1m
  scope: per-account
  source: https://parallel.ai/pricing
  note: Published on the pricing page; not yet listed in the docs rate-limit table.
response_headers:
  published: false
  observed: []
  detail: >-
    Parallel documents no RateLimit-* / X-RateLimit-* response headers and the OpenAPI
    declares none, so an agent cannot read remaining quota from a response. This is the
    material gap - the numbers are published but the runtime signal is not.
exhaustion:
  status_code: 429
  retry_after: undocumented
  note: >-
    A 429 rate-limit status is the conventional exhaustion code and appears in the API
    error reference; no Retry-After semantics are documented.
increases:
  process: Contact support@parallel.ai with use case and requirements.
  contact: support@parallel.ai
free_tier:
  detail: >-
    The hosted Search MCP (https://search.parallel.ai/mcp) is free anonymously at
    "lower rate limits" - the anonymous ceiling is not numerically published. Passing a
    Parallel API key as a Bearer token raises it. Free-tier requests are correlated by
    the `session_id` argument the MCP tools require.
notes: >-
  Limits read verbatim from the published rate-limit reference and cross-checked against
  the pricing page. See conventions/parallel-web-systems-conventions.yml for the wider
  runtime semantics and errors/parallel-web-systems-problem-types.yml for the error envelope.