[Bug]: How to calculate the cost of llm api token used in headless mode

[kevin-support-bot]

All-Hands-AI#7649 Issue

---

@krishgcek, Would you run with DEBUG=1 environment variable?

[Rahulg-ai]

So, to figure out how much you're spending on tokens in headless mode, just check the usage field in the API response — it tells you how many prompt and completion tokens were used.

Then, look up the token pricing for the model you're using (like GPT-4 or GPT-3.5). Each one charges a certain amount per 1,000 tokens for both input and output.

Now do some quick math:
(prompt tokens / 1000 * prompt price) + (completion tokens / 1000 * completion price) — and that’s your cost for that call.

If you're doing this often, just log the usage automatically and have a script add it all up for you.