OpenAI GPT-4 Token Cost Calculator

1. Pick a model and a context size. The tool loads the input, output, cached input and cache write prices from the official list. You can overwrite any of them.
2. Enter input tokens and output tokens per request and your requests per month. The usage object in each API response reports the real counts.
3. Optionally set a cached share and hit rate, or switch to Batch or Flex. Read the monthly cost, the split between input and output, and the saving.

Here is what each result means:

OpenAI token cost is the amount the OpenAI API charges for the tokens in a request and its response. The price list quotes every text model in US dollars per 1M tokens, with separate prices for input, cached input and output.

A token is a chunk of text. OpenAI describes tokens as commonly occurring sequences of characters and gives a rough rule of thumb: 1 token is approximately 4 characters or 0.75 words of English text. That rule is an approximation. The count in the API response is what gets billed.

Is GPT-4 Still on the Price List?

Yes, for a short time. On 2026-10-02 the price list still shows gpt-4-0613 at $30 input and $60 output per 1M tokens, and gpt-4-turbo-2024-04-09 at $10 and $30. The OpenAI deprecations page gives October 23, 2026 as the shutdown date for both and names gpt-5.6-sol as the substitute. This calculator covers the whole current list, so you can price a move away from GPT-4.

This page is an independent estimator and is not affiliated with OpenAI.

Each request is billed as input tokens times the input price plus output tokens times the output price, divided by one million.

1. Multiply input tokens by the input price per 1M tokens.
2. Multiply output tokens by the output price per 1M tokens.
3. Add both and divide by 1,000,000.
4. Multiply by requests per month. Multiply by 0.5 for Batch or Flex processing.

When part of the prompt repeats, that part is priced as cached input. The tool splits the cached tokens by your hit rate: cached tokens x (hit rate x cached price + miss rate x cache write price).

Official OpenAI Text Model Prices, Read 2026-10-02

Standard prices in USD per 1M tokens for requests up to 272,000 input tokens. "Not listed" means the price list shows no figure in that column.

The picker holds more models from the same list. The list notes that the gpt-5.6-sol price is promotional and available at least through November 21, 2026.

An app sends 1,500 input tokens and receives 400 output tokens per request, 200,000 times a month.

On gpt-5.6-sol ($4 and $20): input = 1,500 x $4 / 1,000,000 = $0.006. Output = 400 x $20 / 1,000,000 = $0.008. One request costs $0.014, the month costs $2,800 and the year $33,600. Output tokens are 57.1% of the bill.

On GPT-4, gpt-4-0613 ($30 and $60): input = $0.045, output = $0.024, so one request costs $0.069 and the month costs $13,800. That is about 4.9 times the gpt-5.6-sol bill for the same tokens.

Add caching on gpt-5.6-sol: 60% of the input repeats and 90% of requests hit the cache. The blended cached price is 0.9 x $0.40 + 0.1 x $5 = $0.86 per 1M. One request falls to $0.011174 and the month to $2,234.80, a 20.2% saving.

The Model

Prices on the current list span a wide range. Input runs from $0.05 per 1M tokens on gpt-5-nano to $30 on gpt-4-0613. Pick the model first, because nothing else moves the bill as much.

Output Share

Output costs more than input on every text model in the table above. The ratio is 5 to 1 on the gpt-6 models and 8 to 1 on gpt-5. Reported output includes all generated tokens, not only the visible text.

Cached Input

The cached input price differs by model. It is 0.05 times the input price on gpt-6.1-sol, 0.1 times on most gpt-5 and gpt-6 models, 0.25 times on gpt-4.1 and 0.5 times on gpt-4o. For GPT-5.6 and later, a cache write costs 1.25 times the input price. OpenAI sets the minimum cacheable prompt at 1,024 tokens for those models.

Batch and Flex

The Batch API gives a 50% discount and completes each batch within 24 hours. Flex processing is priced at Batch rates in exchange for slower responses. Neither suits a live chat.

Long Context

Requests above 272,000 input tokens use long-context prices on the models that list them. On those models the long-context input price is double and the output price is 1.5 times the standard price.

Fast Mode and Regional Processing

The list also shows Fast prices above the standard ones, and a 10% uplift for regional processing on models released on or after March 5, 2026. Type those prices into the fields to model them.

GPT-4 is the most expensive general text model left on the list. This table prices one workload on six models: 1,500 input tokens, 400 output tokens and 200,000 requests a month at standard prices.

Price is one side of the choice. Quality on your own tasks is the other, and this page does not measure it. Token counts can also differ between model generations. To express any gap as a percent, use the percentage calculator.

Planning a Move Off GPT-4

GPT-4 snapshots shut down on October 23, 2026. Load the GPT-4 preset, note the cost, then switch the model to see the bill on the substitute.

Setting a Budget

Turn expected traffic into a monthly figure before launch. Compare that figure with the value of the feature in the ROI calculator, and add it to your other subscriptions in the AI tool stack cost calculator.

Checking a Cost Cut

Test an idea before you build it. Enter a shorter prompt, a cached share or Batch mode and read the saving. For reply speed instead of cost, see the token generation speed simulator.

1. Pricing GPT-4 Without a Retirement Plan

The GPT-4 and GPT-4 Turbo snapshots have a published shutdown date. A budget built on them needs a second line for the replacement model.

2. Assuming One Cache Discount for Every Model

The cached input price ranges from 0.05 to 0.5 times the input price. A flat 50% assumption overstates the cached cost on most current models.

3. Ignoring the Cache Write Price

On GPT-5.6 and later, the first request pays 1.25 times the input price for the cached part. OpenAI notes that writing a prefix once and reusing it once costs 1.35 times its normal input cost, against 2 times without caching.

4. Counting Only the User Message

Instructions, tool schemas and chat history are input tokens on every call. Trimming 800 tokens from a prompt sent 200,000 times a month saves $640 at $4 per 1M.

5. Missing the 272K Step

A request with 300,000 input tokens and 2,000 output tokens costs $2.46 at long-context gpt-5.6-sol prices. At the standard prices it would look like $1.24.

6. Estimating Tokens from Characters for Files

OpenAI says estimates like characters divided by 4 are inaccurate for images and files. Use the token counting endpoint for those.

The arithmetic is exact for the prices and token counts in the fields. Real bills differ when real requests differ from your averages.

GPT-4 (gpt-4-0613) is listed at $30 per 1M input tokens and $60 per 1M output tokens on the official price list read on 2026-10-02. GPT-4 Turbo is listed at $10 and $30.

Yes. The OpenAI deprecations page lists October 23, 2026 as the shutdown date for the gpt-4-0613 and gpt-4-turbo snapshots and names gpt-5.6-sol as the substitute model.

OpenAI gives a rough rule of thumb of 0.75 words, or about 4 characters, per token for English text. That puts 1,000 words near 1,333 tokens. It is an approximation, so use the token counting endpoint for a real count.

On the list read on 2026-10-02, gpt-5-nano has the lowest input price at $0.05 per 1M tokens and an output price of $0.40. The cheapest fit for you is the lowest-priced model that passes your own tests.

The Batch API gives a 50% discount compared with standard requests. The trade-off is time: OpenAI says each batch completes within 24 hours, so it suits offline jobs.

Yes. Reused prompt prefixes are billed at the cached input price, which the list sets at 0.05 to 0.5 times the input price depending on the model. Caching is on by default for supported models.

The price list sets a higher output price for every text model, from 2 times input on GPT-4 to 8 times on gpt-5. OpenAI does not publish a cost breakdown, so this page reports the ratio and not a reason.

This page prices API tokens only. The API price list is separate from ChatGPT plans, so check your API billing settings for what your account is charged.

Request sizes vary, cache hits depend on timing, and tools such as web search add separate charges. Compare the token totals on your usage dashboard with the numbers you entered here.

No. The maths runs in your browser and nothing is sent to us. Save keeps a result only in this browser.

• OpenAI API pricing: standard, batch, flex and long-context prices (OpenAI documentation, read 2026-10-02).
• Prompt caching guide: cached input, cache writes and the 1,024-token minimum (OpenAI documentation, read 2026-10-02).
• Batch API guide: 50% discount and 24-hour completion (OpenAI documentation, read 2026-10-02).
• Deprecations: shutdown dates for GPT-4 snapshots (OpenAI documentation, read 2026-10-02).
• Key concepts: tokens and the 4-character rule of thumb (OpenAI documentation, read 2026-10-02).