Billing & Support
Usage-Based Billing
Understand input and output token billing and estimate and control costs from actual usage.
APIQIK charges for actual usage. Input and output rates can differ by model; current model-catalog and console prices are the source of truth.
Billing components
| Component | Field | Description |
|---|---|---|
| Input tokens | usage.prompt_tokens | System instructions, message history, user input, tool definitions, and other input. |
| Output tokens | usage.completion_tokens | Generated text, reasoning, or tool-call content according to the model's billing definition. |
| Total tokens | usage.total_tokens | Usually combined input and output tokens, used for usage monitoring. |
Cost formula
Per-request cost: Input tokens × input rate + output tokens × output rate. Rate units and settlement precision follow the model catalog.
Do not budget from average requests alone. Measure P50 and P95 input/output token counts and daily volume, then reserve capacity for traffic changes, retries, and new features.
Record response usage
response = client.chat.completions.create(
model="qwen-turbo",
messages=[{"role": "user", "content": "Summarize this text"}],
)
usage = response.usage
print({
"prompt_tokens": usage.prompt_tokens,
"completion_tokens": usage.completion_tokens,
"total_tokens": usage.total_tokens,
})Reduce costs
- Match models to task difficulty rather than using the most expensive model for every request.
- Remove unrelated history, repeated system instructions, and retrieval noise.
- Use
max_tokensto prevent unnecessarily long answers. - Cache stable, repeated results where permitted.
- Attribute usage by tenant, feature, and environment to detect unusual growth.
- Apply deterministic filters before generation so models do not handle tasks that require no reasoning.