Azure OpenAI cost: per token, or provisioned throughput units

Quick answer

Azure OpenAI bills standard per 1,000 tokens (input and output priced by model) for pay-as-you-go, or per Provisioned Throughput Unit (PTU), reserved capacity billed hourly (or via a monthly/annual reservation) for predictable, high-volume, low-latency workloads. Standard suits variable usage; PTUs suit steady production at scale. Concise prompts, capped outputs, and the right model are the standard-cost levers.

Azure OpenAI Service provides OpenAI models through Azure. Its pricing mirrors the token model of generative AI, with two consumption options: standard pay-as-you-go per token, and Provisioned Throughput Units for reserved capacity.

Standard: pay per token

On the standard option you pay per 1,000 tokens, with separate input and output rates that vary by model. It has no commitment, so it suits variable, exploratory, or lower-volume usage. Cost is token count times rate, so long prompts, large context, and verbose outputs each add up on every call.

Provisioned Throughput Units

Option	Bills for	Best when
Standard (pay-as-you-go)	Per 1,000 tokens by model	Variable, exploratory, low volume
Provisioned Throughput Units	Per PTU (hourly or reserved)	Steady, high-volume, latency-sensitive
Batch	Discounted for asynchronous jobs	Large offline workloads

PTUs reserve dedicated model throughput, billed hourly or discounted through a monthly or annual reservation. They give predictable performance and cost at high, steady volume, and are cheaper per token than standard once utilization is high. Batch offers a discount for asynchronous jobs.

The token levers

On the standard option, reduce token volume: trim system prompts and context, cap output length, retrieve only relevant context rather than stuffing everything, cache responses for repeated inputs, and choose the smallest model that meets the quality bar. These cut cost on every request, the same discipline as Bedrock.

Choosing between standard and PTUs

Start on standard for variable or growing usage. Move to PTUs when volume is steady and high enough that reserved throughput beats per-token pricing and you need predictable latency and cost. Model the crossover from your token volume before committing to PTUs, so you reserve capacity you will actually use rather than stranding it, the same commitment discipline as any reserved capacity.

FAQ

How is Azure OpenAI priced?

Standard pay-as-you-go per 1,000 tokens, with input and output rates that vary by model, or per Provisioned Throughput Unit (PTU) for reserved capacity billed hourly or via a monthly or annual reservation. Standard suits variable usage; PTUs suit steady, high-volume, latency-sensitive production.

What are Provisioned Throughput Units in Azure OpenAI?

Reserved units of dedicated model throughput, billed hourly or discounted through a monthly or annual reservation. They provide predictable performance and cost at high, steady volume and are cheaper per token than standard once utilization is high. They suit production workloads with consistent demand.

How do I reduce Azure OpenAI cost?

On standard pricing, trim prompts and context, cap output length, retrieve only relevant context rather than stuffing everything, cache responses for repeated inputs, and choose the smallest model that meets the quality bar. For steady high volume, PTUs beat per-token pricing.

Standard or PTUs for Azure OpenAI?

Use standard pay-as-you-go for variable, exploratory, or growing usage with no commitment. Move to PTUs when token volume is steady and high enough that reserved throughput is cheaper per token and you need predictable latency. Model the crossover before committing so you do not strand reserved capacity.

Does model choice affect Azure OpenAI cost?

Yes, substantially. Larger, more capable models cost more per token than smaller ones, sometimes by a large margin. Selecting the smallest model that meets your quality requirement, rather than the most capable by default, is a major cost lever on standard pricing.

Does C3X estimate Azure OpenAI cost?

Azure OpenAI cost is driven by token volume and model choice, which are usage inputs. C3X prices the surrounding Azure infrastructure, and you model token volumes per model to estimate the generative-AI charges.

What to do next

Estimate the infrastructure around your AI workload before you build it. C3X reads your Terraform and prices your resources against a live catalog. Start with the quickstart.