Introduction
Give GPT-5.4 the rows and formulas behind a warehouse report. It can trace a surprising percentage back to the unit used in the denominator.
Key features & highlights
A dashboard may divide returned items by shipped orders. Both columns are counts, but the quotient does not measure the fraction of items returned. Include cell references so a reviewer can reproduce the calculation.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Include the column headings and formulas with the sample rows. at least 1 items · at most 256 items |
max_tokens | No | Leave room for the cell address beside each finding. |
reasoning_effort | No | Start with none for a simple ratio check. Increase effort for interdependent formulas. |
temperature | No | Set temperature only while reasoning is set to none. |
top_p | No | Use a sampling cutoff only for the non-reasoning review. |
tools | No | Give the lookup a sheet name and cell address. at most 128 items |
parallel_tool_calls | No | Independent cells can be retrieved together. Default: |
tool_choice | No | Require the lookup when the denominator is outside the supplied range. |
response_format | No | Request separate fields for the formula, observed unit and proposed correction. |
Pricing
Input $2.50 / Output $15 per 1 million tokens; Prompts over 272K tokens: Input $5 / Output $22.50
Use cases
Send the column headers, the unit for each quantity and a few rows from the reporting period. Ask for separate findings when a formula mixes units and when a record lacks a shipment count.
Prompt guide & pro tips
Register a lookup tool if the model needs more rows. The application should return the displayed value, formula and unit together so the review can distinguish a rounded display from a calculation error.
A screenshot may show the percentage but hide the referenced cells. Include the underlying formula or ask the reviewer to flag the missing denominator.
API quick start
curl --fail-with-body -sN https://api.spicyapi.ai/v1/chat/completions \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
-d '{
"model": "openai/gpt-5.4/chat",
"messages": [
{
"role": "user",
"content": "Summarise the following release notes into three bullet points."
}
],
"stream": true
}' \
| jq --unbuffered -Rrj 'select(startswith("data: ")) | .[6:] | select(. != "[DONE]") | fromjson | if .error then "\n[error] \(.error.message)\n" else (.choices[0].delta.content? // empty) end'
See the API reference for authentication, the complete schema, polling, and error handling.

















