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: Hailuo AI / MiniMax
providerId: hailuo-ai
created: '2026-07-11'
modified: '2026-07-11'
reconciled: false
tags:
- Video Generation
- AI Video
- Generative AI
- Rate Limiting
- Quotas
description: >-
  MiniMax does not publish a single fixed numeric rate-limit table for its
  developer APIs. Video generation is asynchronous and effectively governed by
  concurrent-task limits and your account's video-point balance rather than a
  per-minute request cap: you create a task, then poll GET /query/video_generation
  (MiniMax recommends about a 10-second polling interval to avoid unnecessary
  load). Text/chat and text-to-speech APIs are governed by per-account
  requests-per-minute and tokens/characters-per-minute quotas that vary by model
  and account tier. Concrete numeric limits should be confirmed in the console.
notes: >-
  Poll video tasks roughly every 10 seconds. Generated video download URLs are
  temporary (valid about 9 hours / 32,400 seconds), so retrieve and persist
  assets promptly. Numeric per-endpoint limits are not fully documented publicly
  and differ by model, region (api.minimax.io vs api.minimaxi.chat), and tier.
sources:
- https://platform.minimax.io/docs/guides/video-generation
- https://www.minimax.io/news/video-generation-api
- https://platform.minimax.io/docs/api-reference/speech-t2a-http
responseCodes:
  throttled: 429
limits:
- name: Video Generation Concurrent Tasks
  scope: account
  metric: tasks
  limit: not published (tier-dependent)
  notes: Async video tasks are bounded by concurrent-task limits and video-point balance rather than a request-rate cap.
- name: Video Task Polling
  scope: account
  metric: requests
  limit: recommended ~1 request / 10 seconds per task
  notes: MiniMax recommends a ~10-second polling interval on GET /query/video_generation.
- name: Chat / LLM Requests
  scope: account
  metric: requests
  limit: per-model RPM (not published)
  notes: Requests-per-minute and tokens-per-minute quotas vary by model and account tier.
- name: Text-to-Speech Requests
  scope: account
  metric: requests
  limit: per-model RPM (not published)
  notes: T2A HTTP and WebSocket are bounded by per-account concurrency and character throughput.
- name: Download URL Validity
  scope: file
  metric: seconds
  limit: 32400
  notes: Retrieved video download URLs expire after about 9 hours; persist assets promptly.
policies:
- name: Asynchronous Backpressure
  description: Video generation is queued (Queueing/Preparing/Processing); throughput is managed by the task queue rather than by rejecting requests outright.
- name: Backoff Strategy
  description: Clients should poll on the recommended interval and implement exponential backoff with jitter, honoring Retry-After on 429 responses.
- name: No Charge on Failure
  description: Failed generations and clips blocked by security review do not consume video points.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com