🟢 ChatGPT Token Calculator
What a prompt costs on OpenAI's models. OpenAI prices the same model four ways — Standard, Flex or Batch, Fast and Ultrafast — and the gap between the cheapest and dearest tier of one model is six times, so the tier matters as much as the model.
OpenAI's four-tier pricing, and why the model name is only half the answer
OpenAI prices most models four ways. The same model can be bought at Flex or Batch rates, Standard, Fast, or Ultrafast, and the range between the cheapest and dearest tier is six times. Choosing the tier is therefore as consequential as choosing the model, and it is the part most cost comparisons leave out entirely.
Current prices
| Model | Input | Output | Cache read |
|---|---|---|---|
| GPT-6 Astra | $10 | $50 | $1 |
| GPT-6.1 Sol | $2 | $10 | $0.1 |
| GPT-6 Luna | $0.1 | $0.5 | $0.01 |
| chat-latest | $5 | $30 | $0.5 |
| GPT-5.3 Codex | $1.75 | $14 | $0.175 |
| GPT-5.6 Sol | $4 | $20 | $0.4 |
What the tiers mean
Flex and Batch cost half the standard rate and trade latency for it: results come back more slowly, or in bulk within hours. Standard is the default. Fast and Ultrafast cost two and six times standard respectively, buying priority at times of load. Anything processed overnight, in bulk, or without a human waiting belongs on Flex or Batch, and moving it there is usually a one-line change.
Cached input is where the real discount is
OpenAI charges cached input at a small fraction of the standard rate — on GPT-6.1 Sol it is $0.10 against $2.00, a 95% saving on any part of the prompt that repeats. For an assistant with a long fixed system prompt, that single setting matters more than the choice between adjacent models.
chat-latest is a moving target
The chat-latest alias tracks whatever model currently powers ChatGPT. That is convenient for staying current and awkward for budgeting, because both the behaviour and the price can change without you doing anything. For anything where cost predictability matters, pin a specific model version instead.
Picking the right tier for the job
Astra is the flagship and costs a hundred times Luna on input. Very little production traffic needs a flagship: classification, extraction, routing and short summaries run perfectly on the cheapest tier. The usual pattern is to run Luna first, validate the output against a simple rule, and escalate only the failures to Sol or Astra.
Prices come from OpenAI API pricing and were checked on 2026-10-26. AI pricing changes frequently — confirm against the provider before committing to a budget.
If this is one step in a larger decision, the ai token calculator is usually the next question, and the claude token cost is worth a look while you are here.
Frequently Asked Questions (FAQs)
How much does the OpenAI API cost?
It depends on the model and the tier. On 2026-10-26 the range ran from $0.10 per million input tokens on the lightest tier to $60 on Ultrafast flagship, with output between $0.50 and $300.
What is the difference between Flex, Standard, Fast and Ultrafast?
Priority and price. Flex and Batch are half the standard rate with slower delivery, Fast is double, and Ultrafast is six times. The model and its quality are identical; you are buying queue position.
How much do cached tokens cost?
A small fraction of the standard input rate — on GPT-6.1 Sol it is $0.10 against $2.00, a 95% discount. The saving applies to the repeated portion of a prompt, so it matters most with long system prompts.
Is GPT billed differently for ChatGPT and the API?
Yes. A ChatGPT subscription is a flat monthly fee with usage limits; the API bills per token with no floor. This calculator models the API, which is the one where prompt length shows up on the bill.
What does chat-latest cost?
$5 input and $30 output per million on 2026-10-26, but it is an alias that tracks whatever model powers ChatGPT, so both its behaviour and its price can change without notice. Pin a version if you need stable costs.
How do I cut my OpenAI bill?
In order of impact: move non-urgent work to Flex or Batch for 50% off, enable prompt caching for up to 95% off repeated input, and route simple tasks to a cheaper model rather than defaulting to the flagship.
Are these prices current?
They were taken from OpenAI's published pricing page on 2026-10-26. OpenAI reprices and retires models regularly, so check the official page before finalising a budget.