Introduction
GPT-5.4 Mini examines short bug reports, reads screenshots alongside source excerpts, and returns the next diagnostic step as text or a function call.
Key features & highlights
Give the model the assertion, the failing input and the two functions involved. Ask which observation would distinguish a conversion error from stale data. It can suggest a check without access to the whole repository.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Send the failing assertion before the log excerpt. at least 1 items · at most 256 items |
max_tokens | No | Leave enough answer space for a diagnosis and its supporting observation. |
reasoning_effort | No | Mini defaults to none; choose low, medium, high or xhigh for a harder investigation. |
temperature | No | Set temperature with reasoning turned off for Mini. |
top_p | No | Adjust the sampled probability range only at the none effort level. |
tools | No | Describe the test lookup as a function with named arguments. at most 128 items |
parallel_tool_calls | No | Disable parallel calls if the second check depends on the first. Default: |
tool_choice | No | Require a lookup only when the next step needs data from your runner. |
response_format | No | Use a JSON schema to keep the suspected cause and proposed test separate. |
Pricing
Input $0.60 / Output $3.60 per 1 million tokens
Use cases
Pair the visible error with the relevant handler. Mark what the screenshot actually shows, and ask the model to keep its explanation separate from evidence about the request path.
Prompt guide & pro tips
Define a function that looks up one test by name. Validate the returned arguments in your application, run the lookup there, then return its result so Mini can revise the diagnosis.
The context window holds 400,000 tokens, but the input accepts at most 272,000. Keep the failing case and its dependencies together, and send the rest only if the investigation needs it.
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-mini/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.

















