We may earn a commission if you buy through links on this page. It never changes a score or a ranking. How we make money
Pricing is quote-based.
Azure OpenAI Service has no published starting price in the pricing information we reviewed. Standard (On-Demand) and Batch API charge per token, but their rates are not listed; Provisioned (PTUs) bills per PTU-hour, also without a published rate. A general Azure Free Account offer includes 30 days and $200 in Azure credit, but that is not a dedicated Azure OpenAI free plan. The key budget risk is variable usage, plus undisclosed charges if provisioned traffic spills over to Standard.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Standard (On-Demand) | — | Quote | Variable, consumption-based workloads |
| Provisioned (PTUs) | — | Quote | Workloads needing provisioned throughput |
| Batch API | — | Quote | Non-urgent batch workloads |
Add-ons and overages
- Monthly and annual reservations are available for Provisioned throughput; prices are not published.
- Provisioned traffic that spills over is billed at Standard pay-as-you-go rates; overage prices are not published.
02What the plan grid doesn't tell you
No published token rates
Standard and Batch are billed per token, but their rates are not published here, so you cannot calculate a bill from the plan grid alone.
PTU reservation pricing
Monthly and annual Provisioned reservations are available, but their prices are not published.
Provisioned spillover
Traffic beyond provisioned throughput may be throttled or spill over to Standard. Spillover uses Standard pay-as-you-go rates, which are not published.
Quota and batch limits
Standard uses token-per-minute and request-per-minute quotas, and excess requests are throttled. Batch quota limits apply, with completions returned within 24 hours.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Variable consumption workloads | Standard (On-Demand) | Pay-as-you-go token billing fits variable use, but rates are not published and excess requests can be throttled. |
| Throughput you need to provision | Provisioned (PTUs) | Throughput depends on allocated PTUs; monthly and annual reservations are available, but prices are not published. |
| Non-urgent batch workloads | Batch API | The API lists a 50% discount on Global Standard pricing and returns completions within 24 hours; batch quotas apply. |
04Is Azure OpenAI Service worth the money?
Azure OpenAI Service may suit teams already building in Azure: Microsoft lists it within Foundry Models, alongside tools for model discovery and deployment. But the pricing information leaves out the rates needed to forecast spend. Standard and Batch are usage-based, while PTUs add reservation choices without published prices. Batch has a listed 50% discount on Global Standard pricing, but that still does not reveal the bill. We would compare actual workload estimates before choosing it on price.
Azure OpenAI Service pricing FAQ
How much does Azure OpenAI Service cost?
A starting price is not published in the pricing information reviewed. Standard (On-Demand) and Batch API bill per token, while Provisioned (PTUs) bills per PTU-hour. Their rates are not listed, so the cost depends on usage and cannot be calculated from the plan names alone.
Does Azure OpenAI Service have a free plan or trial?
The pricing information lists no free plan. It notes a general Azure Free Account offer with 30 days and $200 in Azure credit. That offer is for Azure generally, not a dedicated Azure OpenAI Service plan, so check the offer terms before budgeting around it.
What is the difference between Standard and Provisioned?
Standard (On-Demand) uses pay-as-you-go input and output token billing, with quotas measured in tokens per minute and requests per minute. Provisioned (PTUs) bills per PTU-hour, and throughput depends on allocated PTU units. Monthly and annual reservations are available for Provisioned.
Is Batch API cheaper than Standard?
The pricing information lists Batch API at a 50% discount on Global Standard pricing, with completions returned within 24 hours. It is intended for non-urgent batch workloads, and batch quota limits apply. The underlying token rates are not published, so the discount does not provide a specific cost estimate.