How to cut your Azure OpenAI API bill

Mode

Both: automatic caching by default, plus explicit breakpoints you control

Automatic (implicit) prompt caching on by default for supported models; explicit cache breakpoints on GPT-5.6+ Standard pay-as-you-go deployments (not PTU-M).

Cache lifetime

GPT-5.6+: prompt_cache_options.ttl 30m minimum (only value). Earlier models: in-memory typically cleared within 5 to 10 minutes of inactivity, always within one hour of last use; extended retention up to 24 hours on listed models.

Minimum prompt

1,024 tokens, first 1,024 identical

Cache write cost

Models before GPT-5.6: no extra charge to write. GPT-5.6 and later: cache writes can incur charges in addition to discounted reads.

Cache read cost

Discount on input token pricing for Standard deployments; up to 100% discount on input tokens for Provisioned deployments

Storage cost

Not stated

How to enable

On by default. On GPT-5.6+ use prompt_cache_options.mode and add prompt_cache_breakpoint to a content block; reuse a prompt_cache_key.

Request fragment

Copy

prompt_cache_breakpoint: { "mode": "explicit" }

From the docs

“For supported models, cache reads are billed at a discount on input token pricing for Standard deployment types and up to 100% discount on input tokens for Provisioned deployment types.”

“Models before the GPT-5.6 family don't charge extra to write to the cache. On GPT-5.6 models and later model families, cache writes can incur charges in addition to discounted cache reads.”

“The only supported value is `30m`, which is also the default. A cached prefix remains eligible for reuse for at least 30 minutes, but the service might retain it longer.”

“The system typically clears caches within 5 to 10 minutes of inactivity and always removes them within one hour of the cache's last use.”

“Extended prompt cache retention keeps cached prefixes active for longer, up to a maximum of 24 hours.”

“A minimum of 1,024 tokens in length.”

“The first 1,024 tokens in the prompt must be identical.”

“Add `prompt_cache_breakpoint: { "mode": "explicit" }` to a supported prompt content block.”

“Prompt caching is enabled by default for supported models.”

Discount

50% off

50% less cost than Global Standard

Window

24-hour target turnaround (completion_window must be 24h); jobs that take longer are not expired and continue until finished or cancelled

Limits

Max 100,000 requests per file; max input file 200 MB (1 GB with bring-your-own-storage); 500 input files without expiration (10,000 with expiration set); queue bounded by enqueued token quota

Endpoint

Global Batch deployment + Batch API (JSONL lines target /v1/responses or chat completions)

Stacks with caching

Not stated: the provider's pages do not say

From the docs

“It processes asynchronous groups of requests with separate quota and offers a 24-hour target turnaround at 50% less cost than global standard.”

“The service aims to process batch requests within 24 hours, but it doesn't expire jobs that take longer.”

“Currently, you must set this value to 24h. Completion within 24 hours isn't guaranteed, and a job can continue until it finishes or you cancel it.”

“| Maximum requests per file | 100,000 |”

“| Maximum input file size | 200 MB |”

“There's no fixed limit on the number of requests you can batch, however, it will depend on your enqueued token quota.”

• Provisioned throughput (PTU) deployments

  “up to 100% discount on input tokens for Provisioned deployment types”

This site does not track per-model Azure OpenAI prices. Use the sources below for the current rate card.