An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
The 429 here is a token-per-minute (TPM) throttle from the AIServices S0 tier's deployment-level rate limit, not a capacity failure. The default quota per deployment is low until you request an increase, and it gets worse fast when multiple callers share one deployment.
Two things to check:
- Look at your deployment's TPM/RPM quota in Azure AI Foundry under Management Center > Quota, and file a quota increase request if you're consistently near the ceiling.
- If more than one app or environment is hitting the same deployment, put a gateway in front of it (Azure API Management, or Foundry's own gateway capability) so you can enforce per-caller token/rate limits and fail over across deployments instead of letting one noisy caller exhaust the shared quota and 429 everyone else.
I wrote up the gateway plus cost/rate-limit pattern in more detail here, including the policy config for token-based limiting and multi-model failover:
Cost-Safe Guardrails for Microsoft Foundry with the Azure AI Gateway