Why is my Azure OpenAI bill so high?
Whether you call OpenAI directly or through Azure, the drivers of a surprising bill are largely the same. Here's what to look at, including the Azure-specific angles.
The universal culprits
Re-sending large prompts without caching, running everything on the top model, unbounded output, duplicate calls, and non-urgent work on the synchronous tier — these dominate most bills regardless of how you access the models.
Azure-specific things to check
Provisioned throughput (PTU) you're paying for but not fully using, deployments left running, and older model versions priced above their current replacement. Match your PTU to real demand and retire idle deployments.
Turn this into your number
Drop your usage export into the free analyzer and see how much of this applies to your account — provable savings separated from estimates. Nothing is uploaded.
Analyze my usage — free