Guide · cost
PTU vs pay-as-you-go for Azure OpenAI
Azure OpenAI is billed either per token (pay-as-you-go) or by reserving provisioned throughput units (PTU). Pay-as-you-go is cheaper and simpler for spiky or low-volume traffic; PTU is cheaper and gives predictable latency once your traffic is steady and high-volume. The break-even depends on your sustained tokens per minute — this guide shows how to work it out before you commit.
- —Pay-as-you-go bills per token with no commitment — best for early-stage or spiky workloads.
- —PTU reserves dedicated capacity for a fixed monthly cost — best for steady, high-volume production traffic.
- —The break-even is a function of sustained tokens per minute; model it before committing to a reservation.
- —PTU also buys predictable latency, which can matter more than raw cost for user-facing apps.
The two Azure OpenAI pricing models
Azure OpenAI gives you two ways to pay:
- Pay-as-you-go — billed per 1,000 tokens, no commitment. You pay only for what you use.
- Provisioned throughput (PTU) — you reserve dedicated capacity (measured in PTUs) for a fixed monthly cost, with predictable throughput and latency.
How to choose
The decision comes down to two questions: how steady is your traffic, and how much does predictable latency matter?
| Factor | Choose pay-as-you-go | Choose PTU |
|---|---|---|
| Traffic pattern | Spiky or unpredictable | Steady and sustained |
| Volume | Low to moderate | High |
| Latency needs | Tolerant | Predictable / low required |
| Stage | Pilot / early | Established production |
Finding the break-even
Estimate your sustained tokens per minute at peak and typical load. Multiply typical usage out to a monthly pay-as-you-go cost, then compare it to the cost of the PTU reservation that would cover your peak. If your steady usage sits comfortably above the break-even, PTU is cheaper and gives you dedicated latency. Below it, pay-as-you-go wins.
A common pattern is to launch on pay-as-you-go, watch real traffic for a few weeks, then move the steady baseline to PTU and let pay-as-you-go absorb the spikes.
Getting the decision right
North Peak Cloud models this for clients as part of AI cost management — so you commit to reserved capacity only when the numbers say it saves money. If you are unsure which side of the line you are on, that is exactly the kind of thing we work out for you.
Asked and answered.
Is PTU cheaper than pay-as-you-go?+
PTU is cheaper once your traffic is steady and high enough that the reserved capacity is more than fully used. For spiky or low-volume workloads, pay-as-you-go is cheaper because you only pay for the tokens you use.
What is a PTU in Azure OpenAI?+
A provisioned throughput unit (PTU) is a unit of dedicated model capacity you reserve for a fixed monthly cost, giving predictable throughput and latency instead of per-token billing.
How do I choose between PTU and pay-as-you-go?+
Estimate your sustained tokens per minute and compare the monthly pay-as-you-go cost at that volume against the PTU reservation cost. If steady usage exceeds the break-even, PTU wins on both cost and latency; below it, stay pay-as-you-go.