Introduction
GPT 6 Luna is an efficient text-output model for frequent, well-scoped work such as classification, extraction and initial triage. It accepts text and image inputs and supports a 1,050,000-token context window, with at most 922,000 input tokens. Begin with the smallest useful task definition and measure its behavior on the items your application will actually encounter.
Key features & highlights
Luna supports none, low, medium, high, xhigh and max reasoning effort, with medium as the default. The none option suits short decisions with clear rules; a reasoning setting can help when several details must be compared. Its output limit is 128,000 tokens, including reasoning, though a compact classification normally benefits from a much smaller explicit allowance. Sampling controls apply only with reasoning set to none.
Function calling lets Luna request a defined operation, such as handing an uncertain item to a review queue with a typed reason. The application checks the requested arguments and authorization before performing that operation, then returns the result under the matching call identifier. Keep the available function set small enough that the intended handoff remains unambiguous.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Provide Luna with category rules or an extraction schema alongside the item to process; distinguish source passages from instructions. at least 1 items · at most 256 items |
max_tokens | No | Limit Luna’s reasoning and visible labels together; a short triage result rarely needs the full 128,000-token ceiling. |
reasoning_effort | No | Luna accepts none, low, medium, high, xhigh and max, default medium; compare boundary-case accuracy before increasing effort. |
temperature | No | Luna temperature applies only when |
top_p | No | Luna |
tools | No | Offer Luna only the functions needed to route an item or request review; validate the proposed arguments in the application. at most 128 items |
parallel_tool_calls | No | Luna may request independent handoffs together; constrain their execution concurrency in your application. Default: |
tool_choice | No | Set Luna’s function selection to auto, none, required or a named handoff function. |
response_format | No | For Luna extraction, choose text, JSON object or JSON Schema output and check each field against the source. |
Pricing
Input $0.10 / Output $0.50 per 1 million tokens; Prompts over 272K tokens: Input $0.20 / Output $0.75
Use cases
For support-ticket intake, give Luna the routing categories and the evidence that separates them. Ask for a category, the relevant phrase and an uncertainty flag. Preserve a review branch for missing information or a ticket spanning several topics, then compare its assigned routes against a labeled sample.
For document-field extraction, list the required fields, accepted types and representation of absent values. Test scanned layouts, alternate date formats and conflicting totals before using the output in a workflow. A well-formed JSON object proves that parsing succeeded; it does not establish that every extracted value matches the document.
Prompt guide & pro tips
Give Luna concise instructions with examples that expose difficult category boundaries, rather than many nearly identical easy cases. Ask it to preserve identifiers exactly and separate a quoted observation from an inferred label. When the input does not support a required answer, allow the result to record that uncertainty.
For repeated requests, keep stable instructions and schemas consistent so reusable prefixes remain useful. Regenerate document updates only when their contents change, and submit independent items with bounded concurrency in your application. Count reasoning in the output allowance when it is enabled. Inputs above 272,000 tokens use the long-prompt pricing tier for the entire request. Compare accuracy and the cost per accepted result before selecting an effort level for a production queue.
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-6-luna/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.

















