OpenAI API pricing: what GPT-6 costs per token and how to pay less

OpenAI API pricing is pay as you go, in US dollars per million tokens, and since the GPT-6 generation it has more moving parts than a single price per model: a cached-input rate, two cache-write rates, a higher rate for prompts over 272,000 tokens and five processing tiers from Batch to Ultrafast. On October 28, 2026, OpenAI's pricing page lists GPT-6 Luna at $0.10 per million input tokens and $0.50 per million output tokens, GPT-6.1 Terra at $1 and $5, GPT-6.1 Sol at $2 and $10 and GPT-6 Astra at $10 and $50. This guide sets out every current price, explains what each modifier does to the bill and works through a real workload on each model.

ChatGPT subscriptions do not cover any of this: OpenAI's help center says API usage "is separate and billed independently" from ChatGPT Plus. For the consumer plans, see our guide to ChatGPT pricing.

OpenAI API prices for the GPT-6 models

Table: OpenAI API prices per million tokens, prompts up to 272,000 tokens (USD, checked October 28, 2026)

OpenAI describes GPT-6 Astra as "our most capable model for the most demanding work" and the one to start with if you are unsure, GPT-6.1 Sol as "near-Astra performance for complex work at a lower cost", GPT-6.1 Terra, released on October 26, as "a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing" and GPT-6 Luna as its most efficient model "for focused, high-volume tasks". All four have a context window of 1.05 million tokens and up to 128,000 output tokens. GPT-6 Sol, the earlier version of Sol, costs the same except for cached input, which is twice as expensive ($0.20 against $0.10).

Three patterns run through the table:

• Output costs five times input on every GPT-6 model.
• Cache reads are 10% of the input price ($1 on Astra, $0.01 on Luna), except on GPT-6.1 Sol and GPT-6.1 Terra, where they are 5% ($0.10 and $0.05). Writing to the cache costs 1.25 times the input price for the 30-minute cache and 2.5 times for the 12-hour cache that OpenAI added on October 20, 2026.
• Batch and Flex cost half of the standard price, and Fast costs double. Fast is the tier OpenAI called Priority until July 30, 2026, when it renamed it.

Older models are still listed. Among them: GPT-5.6 Sol at $4 and $20 (OpenAI notes promotional pricing for it until at least November 21, 2026), GPT-5.6 Terra at $2 and $12, GPT-5.6 Luna at $0.20 and $1.20, GPT-5 at $1.25 and $10, GPT-5 mini at $0.25 and $2, GPT-4.1 at $2 and $8 and GPT-4o mini at $0.15 and $0.60. A newer model is not always more expensive: GPT-6 Luna costs half as much as GPT-5.6 Luna for input and less than half for output.

Long prompts, processing tiers and regions

Prompts over 272,000 tokens cost more. OpenAI splits every GPT-6 price into short context (up to 272,000 input tokens) and long context (above it). In the long tier, input and cache prices double and output costs 1.5 times as much: GPT-6.1 Sol goes from $2 and $10 to $4 and $15, GPT-6 Astra from $10 and $50 to $20 and $75, GPT-6.1 Terra from $1 and $5 to $2 and $7.50, and GPT-6 Luna from $0.10 and $0.50 to $0.20 and $0.75.

Five processing tiers. Standard is the reference price. Batch and Flex are listed at half price; Batch is the asynchronous option for work that can wait. Fast doubles the price for quicker responses. Ultrafast, listed only for GPT-6 Astra and GPT-6.1 Sol, costs six times the standard rate: $60 input and $300 output per million tokens on Astra.

Regional processing adds 10%. Data residency endpoints carry a 10% uplift for models released on or after March 5, 2026, which includes the GPT-6 family. FedRAMP endpoints carry the same 10%.

Reasoning tokens: the hidden part of the output bill

GPT-6 models reason before they answer, and OpenAI's reasoning guide says those reasoning tokens are billed as output tokens even though they are not visible through the API. They also take space in the context window and count toward max_output_tokens. If that limit is reached before the visible answer starts, the response comes back incomplete and, in OpenAI's words, you "could incur costs for input and reasoning tokens without receiving a visible response".

You control this with the reasoning effort setting. GPT-6 Astra and GPT-6.1 Sol accept low, medium, high, xhigh and max; GPT-6 Luna also accepts none. OpenAI says lower effort "favors speed and lower token usage". For cost estimates, count reasoning in the output, measure it on your own prompts and set max_output_tokens high enough to avoid paying for empty answers.

Tools, images, audio and other models

Built-in tools add their own charges, and the tokens they bring in are billed at the model's rates:

• Web search: $10 per 1,000 calls plus the search content tokens.
• File search: $2.50 per 1,000 calls, and storage at $0.10 per GB per day after the first free GB.
• Containers (Code Interpreter and hosted shell): from $0.03 per 20-minute session for 1 GB to $1.92 for 64 GB, billed by the minute with a 5-minute minimum.

Other prices on the same page, per million tokens unless noted: image generation with gpt-image-2.5-sunburst or gpt-image-2.5-flare at $8 for image input and $30 for image output; realtime voice with gpt-realtime-2.1 at $32 input and $64 output for audio; transcription with gpt-transcribe at an estimated $0.0045 per minute; and embeddings at $0.02 (text-embedding-3-small) or $0.13 (text-embedding-3-large). OpenAI also says it is winding down fine-tuning: the platform is closed to new users.

What a real workload costs on each model

Take a support assistant that answers 100,000 questions a month. Each request sends 2,000 tokens of fixed instructions and reference text that can be cached, plus 1,000 new tokens, and gets back 400 tokens including reasoning.

On GPT-6.1 Sol the month costs:

• New input: 100 million tokens × $2 per million = $200
• Cached input: 200 million tokens × $0.10 per million = $20
• Output: 40 million tokens × $10 per million = $400
• Total: $620 a month, against $1,000 with no caching.

Table: monthly cost of the sample workload (our calculation from OpenAI's prices, October 28, 2026)

The figures leave out cache writes, tools, regional processing and tax, and the Batch column only fits work that can wait. They also cannot be compared token for token with other vendors: the same text produces a different number of tokens on each provider's tokenizer, as we explain in what a token is. Our guide to Claude API pricing runs the same workload on Anthropic's models, and the LLM API price comparison, last checked on October 26, puts four providers side by side with a calculator.

How to keep an OpenAI API bill under control

1. Start with GPT-6 Luna, GPT-6.1 Terra or GPT-6.1 Sol and move up to Astra only for the requests where your own tests show the cheaper model failing.
2. Keep the start of every prompt identical so that it is read from cache at 5% to 10% of the input price. Our prompt caching guide shows how to order a prompt.
3. Send work that can wait to Batch and keep prompts under 272,000 tokens where you can, since the long-context tier doubles the input price.
4. Set the reasoning effort per task. Use low or none for extraction and classification and keep high effort for the hard cases.
5. Watch agents. An agent calls the model and its tools in a loop, so one user request can become dozens of billed calls. Cap the number of steps per request in your own code before you ship.

OpenAI changes models and prices often, and this page is checked against its pricing page every month.