USD list prices for Standard (pay per token) and Batch deployments, model version 2026-03-05.
| Deployment | Input | Cached input | Output | Batch in / out |
|---|---|---|---|---|
| Globallist | $2.50 | — | $15.00 | $1.25 / $7.50 |
| Data Zonelist | $2.75 | — | $16.50 | $1.38 / $8.25 |
| Regional | not in our data yet | |||
Source: Azure Retail Prices API, refreshed 15 Oct 2026. Data Zone and Regional prices for eastus2. "list": price not yet confirmed by the weekly refresh; Data Zone list prices are Global + 10%. GPT-5.4 is a reasoning model: its hidden reasoning tokens are billed as output.
Deployment parameters from Microsoft Learn.
"Fully used" is the 1-month Global reservation divided by everything one PTU can process in a month, in normalized tokens. Real deployments are sized for peak and run below that on average.
Prompts over 128K tokens aren't supported on provisioned deployments.
1M requests a month, 1,500 input and 300 output tokens each, peak 3× the average, Global deployment.
| Standard (Global) | $8,250 |
| Batch | $4,125 |
| Provisioned, hourly | $69,350 |
| Provisioned, 1-month reservation | $24,700 |
| Provisioned, 1-year reservation | $20,995 |
PTUs: peak 226,027 normalized tokens a minute → 95 PTUs, about 33% used on average.
Change the numbers in the calculatorOn a Global Standard deployment, $2.50 per million input tokens and $15.00 per million output tokens. Data Zone costs about 10% more and Batch about half. Prices are USD list prices.
At least 15 PTUs on Global or Data Zone (then steps of 5) and 50 on Regional. Each PTU handles 2,400 input tokens a minute, and one output token counts as 6 input tokens.
A fully used Global PTU on a 1-month reservation costs about $2.47 per million normalized tokens, against $2.50 per million input tokens on Standard. Real deployments are sized for peak load and run well below 100% on average, so Standard is usually cheaper unless the load is high and steady.