Free developer tool

API Rate Limit Calculator

Calculate the requests-per-minute and tokens-per-minute capacity a parallel AI workload needs, including token volume, concurrency, and safety headroom.

No account requiredInputs stay in your browser
Capacity model

Enter workload and active limits

Required capacity
144 RPM
360,000 TPM including margin
Fits entered limits
RPM headroom56
TPM headroom140,000
Provider behaviorVaries by model and usage tier
Official sourceOpenAI
Assumes every worker sustains the entered rate and every request uses the entered token estimate. Bursts and separate input/output limiters can fail earlier.

How to use the result

  1. Enter the maximum number of parallel workers and the sustained requests each worker can start per minute.
  2. Add the typical combined input and output tokens per request, then choose safety headroom for bursts and estimation error.
  3. Compare calculated RPM and TPM independently with the exact model, tier, project, and organization limits shown in your provider console.

How to interpret the answer

Capacity is constrained by whichever active limiter is exhausted first. A workload can fit requests per minute and still fail tokens per minute, concurrency, daily quota, or a provider-specific input/output token bucket. Size the queue and retry policy around the narrowest measured dimension.

A passing result means the entered workload fits the documented constraint represented by this tool. It does not guarantee latency, availability, cost, throughput, or acceptance by every upstream and downstream service.

Assumptions and scope

  • Every worker is assumed to sustain the entered request rate.
  • The token estimate is applied to every request before the safety margin.
  • Enter limits from the relevant console or response headers; provider limits depend on tier and model.
  • Short bursts and separate input/output limiters may require more headroom.

Privacy and verification

The calculation runs locally in your browser and does not require an account. Recheck the linked first-party documentation and your live console or response headers before treating the result as a production configuration.

Related

Read the underlying limits