How do I calculate the cost of an LLM request?

#The formula, with a worked example

Gemini 2.5 Flash-Lite: $0.10 in, $0.40 out per 1M tokens
2,000 input tokens and 300 output tokens:
(2000 * 0.10 + 300 * 0.40) / 1,000,000
= (200 + 120) / 1,000,000
= $0.00032

Rates are the published Gemini 2.5 Flash-Lite rates quoted in LLM Model Routing to Keep Inference Costs Predictable, accessed 2026-08-16. Check the provider's pricing page for current numbers.

#What to log

• One row per attempt, not per run, so a fallback or retry is visible and not hidden.
• Columns: run id, account id, task, model, input tokens, output tokens, cached tokens if the provider reports them, computed cost, timestamp.
• Store the rate you used at the time, because prices change and old rows should stay true.
• Total by run id to get cost per job, then look at the most expensive runs, not the average.

Cached input is priced separately by some providers, and audio or batch modes can differ, so keep the rate in a table, not scattered through the code.