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: Prime Intellect
providerId: prime-intellect
created: '2026-05-25'
modified: '2026-05-25'
reconciled: false
tags:
  - Rate Limiting
  - Quotas
  - GPU Compute
  - Inference
description: 'Best-effort summary of Prime Intellect API rate-limiting and quota behavior. The public documentation
  does not publish a comprehensive table of per-tier RPM/TPM values; control-plane and inference APIs use bearer
  tokens with per-key quotas managed in the Prime dashboard. Reserved quotas for GPU pods are governed by the
  marketplace availability service rather than HTTP throttling.'
sources:
  - https://docs.primeintellect.ai/api-reference/introduction
  - https://docs.primeintellect.ai/inference/overview
  - https://docs.primeintellect.ai/sandboxes/overview
responseCodes:
  throttled: 429
  quotaExceeded: 429
  unauthorized: 401
algorithm: token-bucket
notes:
  - Authentication is HTTP Bearer (Authorization header) using API keys created in the Prime Intellect dashboard.
  - GPU pod provisioning is bounded by marketplace availability, account credit/wallet, and team quota — not by
    classical RPM/TPM rate limits.
  - The inference API at api.pinference.ai follows OpenAI-style usage metering with per-request cost reported.
  - Sandbox creation is governed by per-team concurrency limits visible in the Prime dashboard.
limits:
  - surface: Control plane (api.primeintellect.ai)
    rpm: not-published
    tpm: not-applicable
    notes: Per-key quotas managed in dashboard.
  - surface: Inference plane (api.pinference.ai)
    rpm: not-published
    tpm: not-published
    notes: Usage metering returned in response body; cost computed per request.
  - surface: Sandboxes
    concurrency: per-team
    notes: Concurrent sandbox count is per-team and visible in the dashboard.