Usage, cache, and billing
The user is charged at the public rates of the selected scope. The final amount is recorded after the response completes.
01
Usage fields
| Category | OpenAI-style | Anthropic-style |
|---|---|---|
| Input | prompt_tokens / input_tokens | input_tokens |
| Output | completion_tokens / output_tokens | output_tokens |
| Cache read | cached_tokens | cache_read_input_tokens |
| Cache write | cache_creation_input_tokens | cache_creation_input_tokens |
json
{
"prompt_tokens": 1200,
"completion_tokens": 340,
"prompt_tokens_details": {
"cached_tokens": 800
}
}02
Charging rules
- A failed fallback attempt does not charge the balance.
- Cache read/write is charged only when explicitly reported in upstream usage.
- Local estimation of missing usage does not invent cache hits.
- Images use a fixed price per completed generation.
03
Usage ledger
The cabinet shows model ID, endpoint, tokens, cache HIT, rates, cost, status, and request ID.
Open Usage