OpenAI model guide

OpenAI Models and Token Cost Calculator

Compare GPT-6 Astra with the GPT-5.6 family, estimate one request or a month of API traffic, and see which parts of a prompt actually reach the bill. The calculator is for text API planning, not ChatGPT subscription pricing.

Estimate OpenAI cost Count text tokens

Choosing between GPT-6 and GPT-5.6

OpenAI describes these models as supporting text and image input with text output. The four choices below share a large context window, but they serve different budgets and levels of difficulty. The useful question is not simply which model is best. It is which model clears your quality bar at the lowest total cost.

GPT-6 Astra

Demanding end-to-end work

OpenAI's newest flagship for difficult reasoning, coding, research, and document work. Its higher token rates make task-level evaluations important; check API access during the rollout.

1.05M context / 128K output

GPT-5.6 Sol

Hard reasoning and coding

A capable option for professional reasoning and coding below Astra's token rates. The current Standard promotion lowers both input and output costs from Sol's launch prices.

1.05M context / 128K output

GPT-5.6 Terra

Balanced production work

A middle option for applications that need strong intelligence without paying the highest rate on every request. It is a practical place to start for mixed workloads.

1.05M context / 128K output

GPT-5.6 Luna

High-volume, cost-sensitive tasks

The lower-cost member of the family for routing, extraction, classification, and other repeatable jobs that do not need the most capable model each time.

1.05M context / 128K output

Interactive cost estimate

Estimate OpenAI API cost

Enter the tokens in one request and how often it runs. The comparison uses the current text rates stored on this site for OpenAI models.

Request pattern

Input tokens per request

Cached-read tokens

Expected output tokens

Requests per month

Example workloads

Highest monthly estimate

$14,400

GPT-5.4 Pro

Lowest monthly estimate

$81.6000

GPT-5.6 Luna

Estimated standard text cost

Rates are per 1M tokens. Results are sorted from highest to lowest monthly estimate.

GPT-5.4 Pro

Context 1,050K / 128K

$14,400

per month

Input

$30

Cache

not listed

Output

$180

$1.4400 per request

GPT-5.5 Pro

Context 1,050K / 128K

$14,400

per month

Input

$30

Cache

not listed

Output

$180

$1.4400 per request

GPT-6 Astra

Context 1,050K / 128K

$3,680

per month

Input

$10

Cache

Output

$50

$0.3680 per request

Access is rolling out. Check availability for your API account.

GPT-5.5

Context 1,050K / 128K

$2,040

per month

Input

Cache

$0.5

Output

$30

$0.2040 per request

GPT-5.6 Sol

Context 1,050K / 128K

$1,472

per month

Input

Cache

$0.4

Output

$20

$0.1472 per request

Promotional Standard rates, guaranteed at least through Nov 21, 2026.

GPT-5.4

Context 1,050K / 128K

$1,020

per month

Input

$2.5

Cache

$0.25

Output

$15

$0.1020 per request

GPT-5.6 Terra

Context 1,050K / 128K

$816.0000

per month

Input

Cache

$0.2

Output

$12

$0.0816 per request

GPT-5.4 Mini

Context 400K / 128K

$306.0000

per month

Input

$0.75

Cache

$0.075

Output

$4.5

$0.0306 per request

GPT-5.4 Nano

Context 400K / 128K

$83.6000

per month

Input

$0.2

Cache

$0.02

Output

$1.25

$0.008360 per request

GPT-5.6 Luna

Context 1,050K / 128K

$81.6000

per month

Input

$0.2

Cache

$0.02

Output

$1.2

$0.008160 per request

Model	Rates per 1M	Per request	Per month
GPT-5.4 Pro Context 1,050K / 128K	Input $30 Cache read not listed Output $180	$1.4400	$14,400
GPT-5.5 Pro Context 1,050K / 128K	Input $30 Cache read not listed Output $180	$1.4400	$14,400
GPT-6 Astra Context 1,050K / 128K Access is rolling out. Check availability for your API account.	Input $10 Cache read $1 Output $50	$0.3680	$3,680
GPT-5.5 Context 1,050K / 128K	Input $5 Cache read $0.5 Output $30	$0.2040	$2,040
GPT-5.6 Sol Context 1,050K / 128K Promotional Standard rates, guaranteed at least through Nov 21, 2026.	Input $4 Cache read $0.4 Output $20	$0.1472	$1,472
GPT-5.4 Context 1,050K / 128K	Input $2.5 Cache read $0.25 Output $15	$0.1020	$1,020
GPT-5.6 Terra Context 1,050K / 128K	Input $2 Cache read $0.2 Output $12	$0.0816	$816.0000
GPT-5.4 Mini Context 400K / 128K	Input $0.75 Cache read $0.075 Output $4.5	$0.0306	$306.0000
GPT-5.4 Nano Context 400K / 128K	Input $0.2 Cache read $0.02 Output $1.25	$0.008360	$83.6000
GPT-5.6 Luna Context 1,050K / 128K	Input $0.2 Cache read $0.02 Output $1.2	$0.008160	$81.6000

This estimate covers text input, eligible cached reads, and generated text. It does not include cache writes, built-in tools, web search, containers, images, audio, or other separately billed features.

What goes into an OpenAI request

The visible user message is often only a fraction of the input. System and developer instructions, examples, retrieved documents, prior messages, tool definitions, and structured output schemas can all add tokens. If you are estimating a production feature, count the assembled request rather than the text box the user sees.

Normal input

This is new prompt content processed at the standard input rate. Retrieval-heavy applications can spend more here than expected because every attached passage becomes part of the request.

Cached input

Stable prompt prefixes may qualify for a discounted cached-read rate. Keep reusable instructions and examples consistent, but do not assume every repeated-looking prompt will be billed as a cache hit.

Generated output

Output is usually priced above input. A tight response format, a sensible maximum output, and early stopping can matter more than shaving a few lines from a small prompt.

Long context and extras

Some model records use higher rates after a long-context threshold. Tools, web search, containers, images, audio, cache writes, and other features can add charges that a text-only estimate does not capture.

Counting tokens before and after a request

A browser counter is useful while writing a prompt, setting limits, or comparing rough cost. It cannot see every model-specific detail in a fully assembled API call. OpenAI's token-counting guide recommends counting with the model and request shape you plan to use, while production responses provide the usage data you should rely on for billing reports.

That difference matters with chat history, tools, images, and structured messages. Use this page to make decisions early; use provider usage fields to reconcile what actually ran.

Useful checks before launch

Count the final system and developer instructions.
Include retrieved context and tool definitions.
Set a realistic output budget from observed responses.
Track cache hits separately from ordinary input.
Compare estimated totals with API usage in staging.

OpenAI token calculator FAQ

Which OpenAI model should I start with?

Terra is a useful quality and cost baseline. Test Luna for routine, high-volume work and Sol for harder requests. Evaluate Astra when those models fall short, and confirm that your API account has access.

Are GPT-5.6 Sol's lower prices permanent?

OpenAI describes the current Standard rates as promotional and guarantees them at least through November 21, 2026. It has not specified a replacement rate or exact end date, so a longer-term budget should be checked against the official pricing page.

Does cached input reduce every OpenAI request?

No. Cached pricing applies when a request reuses eligible prompt content under OpenAI's caching rules. A changing prompt or a short request may not receive the same benefit.

Is ChatGPT subscription pricing included here?

No. This page estimates API usage from token rates. ChatGPT plans are separate products and are not calculated from the token fields on this page.

Will this browser estimate exactly match my OpenAI bill?

Treat it as a planning estimate. Production usage can include model-specific tokenization, tools, cache writes, media, and other billable items. Use OpenAI's request usage data for final accounting.