Compare API cost with the workload in mind
A useful API price page should explain how to compare models, token mix, retries, tools, and limits. This guide points to live values instead of making a fragile permanent «cheapest API» promise.
What changes your bill
- Input and output tokens, including long context and tool results.
- Model family, capability tier, and any provider-specific multiplier.
- Retries, streaming interruptions, image or audio inputs, and background jobs.
- Minimums, rate limits, account balance rules, and taxes or payment fees where applicable.
A fair GPT versus Claude test
Choose a fixed prompt set from your real application, freeze the model versions, run enough requests to smooth latency variance, and record successful completion rate, p50/p95 latency, input/output tokens, and total spend. Publish the date and methodology when making a public comparison.
Where to check current values
Use live pricing catalog for the live model catalog, interactive setup docs for request examples, and service status for incidents. If you need a security or retention answer before purchase, read security overview and the data-retention page.
Frequently asked questions
Does router.cheap always have the lowest price?
No permanent claim is made. Compare the current catalog with your workload, reliability needs, and the total cost of retries and operational time.
Is there a free API key?
The service uses account-based credentials. Check the current signup and billing terms rather than relying on shared or unofficial keys.
How are tokens counted?
Usage is reported by the API response and account dashboard. Validate the exact model and response fields used by your client.