Token cost across Azure OpenAI models
Same tokens and volume, priced across the catalog (token cost, before overhead) — cheapest first.
| Model | In / Out (per 1M) | Per request | Monthly tokens |
|---|
The token bill is not the Azure bill
The most common Azure OpenAI surprise is an invoice several times larger than the token estimate. It happens because Azure OpenAI is a managed enterprise service, and the tokens are only the metered core. Around them sit costs that a token calculator never shows: a paid support plan (Developer, Standard or Professional Direct, from roughly $100 to over $1,000 a month), private networking and endpoints, data egress, logging and storage, and — if you reserve capacity — idle PTU hours you pay for whether you use them or not. Teams routinely find production Azure OpenAI running 20–40% above the raw token line once all of that is counted. This calculator makes you name that overhead up front and adds it on top, so the number you plan with is the number you're billed.
On the token side, Azure charges exactly like the direct OpenAI API: a per-million rate for input and a higher rate for output. Output dominates — it's priced several times above input on the flagship tiers — so the length of your responses drives the bill more than the size of your prompts. Enter a realistic input and output length and your monthly volume, and the tool separates the token cost from the overhead, so you can see which lever actually moves your total.
Pay-as-you-go vs PTU: where the line is
Azure offers two ways to buy the same models. Pay-as-you-go bills per token — you pay only for what you use, ideal for spiky or low-to-medium volume. A Provisioned Throughput Unit (PTU) reserves a fixed slice of capacity for a fixed monthly price (around $2,000+ per unit) with no per-token charge — ideal when volume is high, steady and latency-sensitive. The trade-off is simple: a PTU is cheaper only once your equivalent pay-as-you-go token bill would exceed the PTU's fixed price. Below that, you'd be paying for capacity you don't use. This tool compares your estimated token cost against the PTU price you enter and tells you which plan wins at your volume — and how far you are from the crossover.
If Azure's overhead is the deciding factor, it's worth pricing the same model on the direct OpenAI API or the AWS Bedrock route, and checking the cheapest LLM API across providers. For the token maths on a single model, the LLM token cost calculator pairs well with this page, and if you're building a product on top, run the margin through the AI wrapper pricing calculator.
How to use it
1. Pick a model and enter your monthly request volume.
2. Enter typical input and output tokens per request.
3. Set the overhead percentage you expect from support, networking and egress (20–40% is common in production).
4. Enter a PTU monthly price to see whether reserving capacity would beat pay-as-you-go at your volume.
Common mistakes
Budgeting the token cost as the total. Support, networking and egress are separate line items — add them as overhead. Reserving a PTU too early. Below the break-even volume you pay for idle capacity; stay pay-as-you-go until usage is high and steady. Estimating on input only. Output is the expensive half; use a realistic response length. Ignoring the support tier. A mandatory paid support plan can exceed a small token bill on its own.
FAQ
Are Azure OpenAI token prices the same as OpenAI's?
Yes, essentially — Azure lists the same per-token input and output rates for the same model. The difference in total cost comes from Azure's surrounding enterprise charges, not the tokens.
What's a realistic Azure overhead percentage?
Production deployments commonly add 20–40% over the token bill for support, networking, egress and monitoring. Light dev usage can be lower; heavily governed enterprise setups can be higher.
How much does a PTU cost?
Roughly $2,000+ per unit per month as a fixed reservation, with no per-token charge. The exact price and how many units you need depend on your throughput and region; treat the field here as an approximation.
Should a startup use Azure OpenAI or the direct API?
If you don't need Azure's compliance, networking or existing Azure billing, the direct OpenAI API avoids most of the overhead. Choose Azure when enterprise governance, data residency or an existing Azure commitment outweighs the extra cost.
Estimate only. Azure OpenAI prices and PTU sizing change — verify current rates with Microsoft before budgeting.