For a standard workflow run using 1,500 input tokens and 500 output tokens, OpenAI API costs range from $0.90 to $56.25 per 1,000 executions. The variance depends entirely on the model tier, with lightweight models like gpt-5.6-luna offering the highest efficiency for high-volume tasks.
Finance and IT leaders must evaluate these unit economics when moving agents from pilot to production. While specialized models like gpt-5.6-cyber provide advanced capabilities, they represent a significant increase in recurring operational expenditure compared to the terra or sol tiers.
What OpenAI API costs per 1,000 workflows
Assumption (stated, adjustable): each workflow run uses 1,500 input tokens and 500 output tokens.
| Model |
Input $/1M |
Output $/1M |
Cost per workflow |
Cost per 1,000 workflows |
| gpt-5.6-luna |
$0.2 |
$1.20 |
$0.0009 |
$0.9 |
| gpt-5.6-sol |
$0.4 |
$20 |
$0.0106 |
$10.60 |
| gpt-5.6-terra |
$2 |
$12 |
$0.0090 |
$9 |
| gpt-realtime-2.1 |
$4 |
$24 |
$0.0180 |
$18 |
| gpt-image-2 |
$8 |
$30 |
$0.0270 |
$27 |
| gpt-5.6-cyber |
$12.50 |
$75 |
$0.0563 |
$56.25 |
Cost per workflow = (input tokens x input price + output tokens x output price) ÷ 1,000,000, using verified official per-token prices.
How do token ratios influence your total cost of ownership?
Every workflow run is billed based on two distinct rates: input tokens for processing instructions and output tokens for generating responses. Because output tokens are priced at a premium—up to six times the rate of input tokens on certain models—workflows that generate long-form content will naturally incur higher costs than those used for simple data extraction or classification.
Why does model selection impact the procurement budget?
The price difference between the most economical and most expensive models is more than 60x. For example, migrating a fleet of 1,000 workflows from gpt-5.6-luna to gpt-realtime-2.1 increases the cost from under a dollar to $18.00. Procurement teams should ensure that the performance requirements of a specific use case actually demand the higher-priced model tiers to avoid unnecessary spend.
To determine if this API spend is a productive investment, you must measure the labor hours reclaimed or the error reduction against the hard cost of the tokens. NetLift provides a framework to track this AI value, mapping specific model costs like the $9.00 per 1,000 workflows for gpt-5.6-terra against your business baseline to ensure a positive net return.
Sources
All pricing on this page comes from official vendor pages: