Service Level Agreement (SLA)

Last updated: 2026-08-16

Availability, dedicated TPM/RPM, support response, incident notice, and service credits are written into an enterprise contract. Peak throughput and measured TTFT are not SLA guarantees. Peak throughput and measured TTFT: /performance.

What this document covers

Developer and Business accounts are best-effort or platform service objectives. Compensable SLA terms apply only when an enterprise contract is signed.

开发者与商务账户为尽力而为或平台服务目标。可追责、可抵扣的 SLA 仅在签署企业合同后生效。

Enterprise availability

  • Monthly API availability target: 99.9% or 99.95%, as written in the contract.
  • Measured on the ModelAPI gateway for signed production endpoints (api.aimodelapi.ai and contracted PoPs).
  • Upstream model-provider or cloud outages are excluded from availability and from service credits.
  • Scheduled maintenance announced on /status is excluded.

Dedicated capacity

  • Paid default after top-up: 500 RPM. 1000+ RPM is requestable, not automatic.
  • Enterprise reserved TPM/RPM is whatever the contract states. We do not SLA-guarantee 200M tokens/min or TTFT P95 under 7s.
  • 200M tokens/min is peak platform capacity and requires upstream quota to be opened in advance.

Support and incidents

  • Navigator in-app support: 7×24 automated assistance for all accounts.
  • Email support@aimodelapi.ai: billing disputes within 24 business hours.
  • Enterprise: named support, incident notice, and response times as contracted.

Service credits

  • Credits (if any) are defined in the enterprise contract: how availability is calculated, the ticket window, and the credit percentage.
  • No public credit table is offered for Developer or Business accounts.
  • Credits are the sole remedy unless the contract says otherwise.

Data and security

Prompts are routed to upstream providers to fulfill your request. We do not use API content to train ModelAPI models. Limited logs are kept for billing disputes, abuse investigation, and legal compliance.

提示词会转发至上游以完成推理。我们不用 API 内容训练自有模型。仅在计费争议、滥用调查和法律要求范围内有限保留。

Exclusions

  • Force majeure, DDoS, or customer-side misconfiguration.
  • Upstream provider outages, model deprecations, or quota the customer did not reserve.
  • Peak marketing figures (200M tokens/min, cache hit up to 98%, measured TTFT) unless explicitly restated as contracted commitments.