generated: '2026-08-14'
method: searched
source: https://docs.parallel.ai/getting-started/rate-limits
also:
- https://parallel.ai/pricing
notes: >-
  Only POST requests that create new resources are rate-limited; GET requests
  (retrieving results, checking status) do not count. Contact support@parallel.ai
  for higher limits. Each limit below names the exact operation it counts, taken
  verbatim from the published table.
limit_count: 7
response_headers:
  documented: false
  note: >-
    Parallel documents the numeric quotas but publishes NO rate-limit response
    headers — no X-RateLimit-*, no RateLimit-*, no Retry-After is described in the
    rate-limit page or the error reference. An agent therefore has no runtime
    signal of remaining budget and can only detect exhaustion by receiving the
    error status. This is the single largest DX gap in Parallel's runtime
    semantics.
exhaustion:
  status: 429
  meaning: Too Many Requests — rate limited or quota exceeded
  retryable: true
  guidance: Implement exponential backoff.
  see: errors/parallel-problem-types.yml
scope: per-account API key, counted per product endpoint; POST-create only
limits:
- endpoint: Search
  operation: POST /v1/search
  limit: 600
  window: minute
- endpoint: Extract
  operation: POST /v1/extract
  limit: 600
  window: minute
- endpoint: Tasks / TaskGroups
  operation: POST /v1/tasks/runs or POST /v1/tasks/groups/{taskgroup_id}/runs
  limit: 2000
  window: minute
- endpoint: Chat
  operation: POST /v1beta/chat/completions
  limit: 300
  window: minute
- endpoint: FindAll
  operation: POST /v1beta/findall/runs
  limit: 300
  window: hour
  discrepancy: >-
    The pricing page (https://parallel.ai/pricing) states a FindAll limit of
    "25 / hr", against "300 per hour" on the rate-limits page. Both are published
    by Parallel and they disagree; the rate-limits page is treated as canonical
    here and the conflict is recorded rather than resolved.
- endpoint: Entity Search
  operation: POST /v1beta/findall/entity-search
  limit: 600
  window: minute
- endpoint: Monitor
  operation: POST /v1alpha/monitors
  limit: 300
  window: minute
  note: >-
    The rate-limit page still names the v1alpha monitor path; the current OpenAPI
    publishes POST /v1/monitors.
free_tier:
  surface: https://search.parallel.ai/mcp
  mechanism: session_id-based free-tier rate limiting; an API key as a Bearer token raises limits.
  docs: https://docs.parallel.ai/integrations/mcp/search-mcp
higher_limits_contact: support@parallel.ai