Introduction
GPT-5.5 compares dated revisions and traces a changed requirement to its source. Return an amendment table with the applicable wording and open questions.
Key features & highlights
Send each version with a date and section identifier. Ask for one row per changed obligation, including the passage that supersedes the original and anything the later text leaves unresolved.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Place revision identifiers next to the passages they describe. at least 1 items · at most 256 items |
max_tokens | No | Allow one answer row per changed section and room for cited wording. |
reasoning_effort | No | GPT-5.5 starts at medium. Choose none for a literal side-by-side extraction. |
temperature | No | Choose none before supplying a temperature for this comparison. |
top_p | No | Set the sampling cutoff only with GPT-5.5 reasoning disabled. |
tools | No | Give the section lookup a revision and a section number as separate arguments. at most 128 items |
parallel_tool_calls | No | Independent section lookups may run in parallel. Default: |
tool_choice | No | Require the lookup on the turn where a passage needs verification. |
response_format | No | Define separate schema fields for earlier wording, replacement wording and source. |
Pricing
Input $4 / Output $24 per 1 million tokens; Prompts over 272K tokens: Input $8 / Output $36
Use cases
A revised packing list may change the number of containers without changing the quantity inside each one. Keep those quantities in separate columns and identify the revision behind each figure.
Prompt guide & pro tips
Let the application supply a lookup function for section identifiers. A matching identifier lets the reviewer retrieve the passage. A missing one leaves the row open for review.
For a standard request above 272,000 input tokens, the input rate doubles and the output rate rises by half across the request. Count both versions when deciding which sections to send together.
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-5.5/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.

















