OpenAI API pricing per model, November 2026

The four flagship models

These are the prices in the Flagship models table of OpenAI's API pricing page as we read it on 12 November 2026. They are in US dollars per 1 million tokens, at standard processing. OpenAI's business pricing page adds that these rates apply to context lengths under 272K.

Output costs five times as much as input on all four models. The largest model, GPT-6 Astra, costs 100 times as much per token as the smallest, GPT-6 Luna.

What 10 million tokens cost

Per-million prices are hard to picture, so here is the same table multiplied by ten: what 10 million input tokens and 10 million output tokens cost on each model at the standard rate, with no caching. A real job is a mix of the two, so read the columns separately and add up your own mix.

Ways to pay less

• Batch: OpenAI's Batch API guide gives a 50% cost discount compared to synchronous requests. The trade is time: each batch completes within 24 hours, and often sooner. It fits work nobody is waiting for.
• Flex: the Flex processing guide says tokens are priced at Batch API rates, in exchange for slower responses and occasional resource unavailability. Flex is in beta with limited model availability, and we did not read which models support it. You are not charged for a request that fails with a resource unavailable error.
• Prompt caching: tokens read from the cache are billed at the cached input rate in the table above, which is a tenth of the input price on Astra and Luna and a twentieth on Sol and Terra. Writing to the cache costs more than plain input: 1.25 times the input rate for a 30-minute cache and 2.5 times for a 12-hour cache, which is offered in the Responses API. For Astra that is $12.50 and $25.00 per million tokens, for Sol $2.50 and $5.00, for Terra $1.25 and $2.50, for Luna $0.125 and $0.25.
• When caching pays off: OpenAI's guide gives the sum for models where reads cost a tenth of input: with a 30-minute cache, a prefix written once and fully reused once costs 1.35 times its ordinary input cost, against 2 times without caching. A 12-hour write costs more, so it needs more reuse to pay off.

What is billed on top of tokens

Tokens are not the whole bill. The developer pricing page lists prices for the built-in tools, and says that tokens used by those tools are billed at the chosen model's per-token rates.

What we could not read

• The exact Batch, Flex, Fast and Ultrafast prices per model. They sit behind switches on the pricing page, and our reading returned the Standard view only. We give the 50% batch discount from OpenAI's guide, not a price per model.
• Prices for long context. The table has a Short and a Long setting. The prices we read match the business page's rates, which it says are for context lengths under 272K. We did not read the Long prices.
• Other models. The page has an "All models" list that our reading did not open. Its notes mention promotional pricing for GPT-5.6 Sol until November 30, 2026, without the price in the text we read.
• Tax, and what you pay through Amazon Bedrock or Microsoft Azure. OpenAI says its models on those platforms are billed through those services.

Knowing what API work cost per task

A price list tells you the rate. It does not tell you what a task cost, because that depends on how many tokens the work used, how much of the input was cached and which model ran it. A price list has no idea which client or task the tokens were for.

Hourtick does not sell OpenAI's models and does not change what you pay OpenAI. It is a time tracker where an AI agent works on a task and reports its own time and model cost on it, in US dollars with the model and the tokens, next to the hours people log. The numbers are what the agent reports. If you connect OpenAI, those reports are checked against the provider's bill every month, and a month where an agent reported more than the bill is flagged.