How the brand sounds (voice)
ai_models · runnable · billing linear
| Fact | Value |
|---|---|
| Provider | ai_models |
| Module | base |
| Runnable today | Yes — a quantity-bound adapter exists |
| API path | /chat/completions |
| Billing kind | linear |
| Price status | Known — a published or measured rate is recorded |
| Availability | Executable in selected plans; target and quantity bounds required |
| Query contract | Enforced — the collector reads these parameters from the contract |
| Catalogue research date | 2026-09-27 |
Billing formula
What the provider charges for one request of this item. Provider cost is not the customer price and is not an account balance movement; the three are different numbers and this is the first of them.
prompt_tokens * input_rate + completion_tokens * output_rate, capped by the contract's input and output limits
Rates
| Component | Value |
|---|---|
input_per_million_usd | 0.25 |
output_per_million_usd | 2.00 |
ceiling_usd | 0.022664 |
Provider limits: {"max_completion_tokens":6000,"max_markdown_bytes":20000,"max_pages":12}
Rate verification: verified_live_2026-09-27.
Caveat: The ceiling counts every UTF-8 byte of the prompt at every limit, plus the response schema and message framing, as an input token (a byte-level tokenizer spends at least one byte per token), and the output at its cap. The charge is settled from the Gateway's own reported cost on the response (measured equal to these rates on 2026-09-27); the pull's Gateway balance check confirms it.
Runtime limits
| Limit | Value |
|---|---|
quantity | 1 |
groups | 1 |
Frozen acquisition parameters
One bounded model call. Every tone claim must quote the page verbatim or it is dropped; the ceiling is the prompt at every limit plus the output cap at the pinned model's published token prices.
| Parameter | Value |
|---|---|
| Fixed parameters | {"model":"openai/gpt-5-mini","max_completion_tokens":6000,"max_markdown_bytes":20000,"max_pages":12,"response_format":"json_schema_strict"} |
| Server-side filters | None |
| Order by | None |
| Page window | 1 |
| Varies by target | pages_read |
Dated sources
https://ai-gateway.vercel.sh/v1/models(read 2026-09-27)
Notes from the catalogue
- A ceiling, not a charge: the receipt records the Gateway's reported charge for the tokens the call used.
- Model pinned in the frozen contract (openai/gpt-5-mini); page text is capped at 20,000 UTF-8 bytes on up to 12 pages and output at 6,000 tokens.