Cost of sustaining N requests per second at your token sizes.
Requests per second is usually discussed for rate limits, but it's also a cost rate. Convert your traffic to a monthly bill and there are no surprises. See limits on the rate limit calculator.
The Concurrent Requests Cost Calculator estimates what a steady stream of API traffic costs over time. You enter your requests per second, the average input tokens and output tokens per request, and your provider's input and output prices per million tokens. From these it projects total token throughput and multiplies it by the per-token rates to show hourly, daily, and monthly spend. The main cost drivers are request rate and token size: doubling your requests per second doubles the bill, and long prompts or verbose responses raise it just as directly, since output tokens are often priced higher than input.
A practical tip is to test each input separately to see which one moves your total most. Trimming prompt context, capping response length, or smoothing traffic peaks usually cuts cost faster than switching providers. The key trade-off is throughput versus spend: higher concurrency serves more users but scales cost linearly, so size your request rate to real demand rather than a theoretical maximum.
Cost per request × requests per second × seconds. For a monthly figure at steady load, multiply by 86,400 (seconds/day) × 30. It answers 'what does this throughput cost to run 24/7?'
Peak RPS is a worst-case ceiling; average load is what you actually pay. Size infrastructure for peak but budget dollars closer to average, since traffic is rarely flat-out all day.
📚 New here? Learn how rate limits & throughput work →