Understanding the dashboard: requests / Tokens / cost
How the metrics differ
| Metric | Meaning | Note |
|---|---|---|
| Requests | How many times the API was called | Streaming output still counts as one request |
| Token | The amount of text the model processes (input + output) | The main billing unit |
| Cost (spending) | The actual charge calculated as Token × unit price | The group multiplier affects the unit price |
| Concurrency | The upper limit on simultaneous requests | Going over it gets you rate-limited (429) |
What to look at on the dashboard
- Usage trend chart: see how Token/request volume changes over time and spot peaks
- Model distribution: which model gets used the most, so you can optimize cost
- Group statistics: how much each group (Claude/GPT/Gemini) consumes separately
- Balance: your real-time balance; you get a reminder when it drops below the threshold (requires enabling balance reminders)
How cost is calculated
Cost = Token consumption × model unit price × group multiplier- The group multiplier is set by the site owner (1.0 = original price, 2.0 = double)
- The amount actually charged in "usage records" is what counts
Common misunderstandings
- Streaming output is not "billed as it generates"; it is billed by the final complete Token count
- A cache hit may not be billed or may be billed at half price (depends on site configuration)
- Image generation is billed per image/by size and does not go through Tokens
TIP
Think the cost looks off? Go to usage records and filter once by key and date; the details of a single request show the input/output Tokens and the charge.