.png){kind=logo}
Token volume, model choice, and output length are the biggest drivers of API costs.
Cached inputs, Batch, and Flex can significantly reduce costs for suitable workloads.
Accurate budgeting requires real usage data, including tools, context size, and service options.
OpenAI API cost depends on model choice, token volume, cache use, context size, and extra tools. A simple request can cost a tiny fraction of a cent, while a large agent task can create a much larger bill.
The current standard rate for GPT-5.6 Sol stands at USD 5.00 per 1 million input tokens, USD 0.50 per 1 million cached input tokens, and USD 30.00 per 1 million output tokens. GPT-5.6 Terra costs USD 2.00 for input, USD 0.20 for cached input, and US$12.00 for output per 1 million tokens. GPT-5.6 Luna costs USD 0.20 for input, USD 0.02 for cached input, and USD 1.20 for output. These rates apply to context lengths under 270K. OpenAI describes Sol as its flagship model, Terra as a balance of quality and cost, and Luna as a low-cost choice for high-volume work.
The model gap can shape a monthly bill. One million input tokens cost 25 times more on Sol than on Luna, and output has the same 25-to-1 gap. Terra sits between the two.
Tokens act as the units that OpenAI counts for text. A token can equal a full word, part of a word, punctuation, or a space. OpenAI gives a simple English rule: one token equals about four characters or about three-quarters of a word. One hundred tokens equal about 75 words, and one paragraph has about 100 tokens. A text of about 1,500 words has roughly 2,048 tokens.
The basic cost formula stays simple. Input token cost equals input tokens divided by 1 million, multiplied by the input rate. Output cost follows the same formula with the output rate. Total model cost equals input cost plus cached input cost plus output cost.
Also Read - OpenAI Responses API: Features, Use Cases, Examples
Repeated context can qualify for cached input rates. GPT-5.6 Sol lists USD 0.50 per 1 million cached input tokens against USD 5.00 for normal input. Terra lists USD 0.20 against USD 2.00, while Luna lists USD 0.02 against USD 0.20.
Large prompts need extra care. OpenAI lists a higher rate for requests above 270K input tokens. For Sol, long-context standard rates rise to US$10.00 per 1 million input tokens, USD 1.00 for cached input, and USD 45.00 for output. Terra rises to USD 5.00, USD 0.50, and USD 22.50. Luna rises to USD 2.00, USD 0.20, and USD 9.00.
A model call does not always represent the full API bill. Web search costs USD 10.00 per 1,000 calls, plus search content tokens at the chosen model rate. File Search costs USD 0.10 per GB per day for storage, with 1 GB free, plus USD 2.50 per 1,000 tool calls. Containers cost USD 0.03 for 1 GB, USD 0.12 for 4 GB, USD 0.48 for 16 GB, and USD 1.92 for 64 GB per 20-minute session.
Batch API offers a 50% discount on input and output for tasks that can wait up to 24 hours. Flex offers lower cost with slower response times and occasional resource limits. Data residency adds 10% for eligible models and endpoints.
Also Read - OpenAI Realtime API: How It Works and When to Use It?
A useful calculator needs four core values: input tokens, cached input tokens, output tokens, and model rate. Extra fields should cover context size, tool calls, service mode, and data residency. A simple estimate can start with 1 million token blocks, then add each cost category.
OpenAI also provides usage and cost tools that show API spend by invoice line item and project. The Costs endpoint can reconcile with the invoice, which makes it suitable for financial checks rather than a rough token estimate.
The safest cost estimate comes from real request data rather than a model rate alone. Token volume, cache share, output length, context size, tool use, and service tier all shape the final bill. A low token rate can still create a high monthly cost when request volume rises. A careful calculator turns those factors into a clear budget before an API workload reaches scale.
What determines OpenAI API cost?
Model, input tokens, output tokens, cached tokens, context size, and additional tools all affect the bill.
How can cached tokens reduce costs?
Repeated context may qualify for lower cached-input rates, reducing the cost of recurring prompts.
Do OpenAI tools have separate charges?
Yes. Tools such as web search, File Search, and containers can add costs beyond model token charges.
Does a larger context cost more?
Yes. Requests exceeding certain context thresholds can have higher token rates.
What is the best way to estimate API expenses?
Use real request data and account for token usage, cache rates, tools, context size, and service tier.
