Introduction
GPT 5.6 Terra occupies the middle tier of the GPT 5.6 range, balancing capability and cost for everyday work. It can read a combination of text and images, return a structured answer and request functions supplied by an application. A 1,050,000-token context allows related material to accompany a task without requiring every request to fill that space.
Key features & highlights
Terra offers six levels of reasoning: none, low, medium, high, xhigh and max. The default medium setting can be adjusted for the problem at hand. A focused extraction may need little deliberation, whereas reconciling inconsistent records benefits from room to examine competing explanations before choosing a result.
The model can take a screenshot, a short specification and the code that implements a feature as one conversation. Function calls let your program supply missing records or validate a proposed action during that discussion. Your program owns execution and authorization, and should return the matching call result so Terra can finish from observed facts rather than an assumed outcome.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Keep the task boundary explicit: name the deliverable, the information that controls it and the conditions that make it acceptable. at least 1 items · at most 256 items |
max_tokens | No | Shift handoff — The output limit covers reasoning and visible text together. |
reasoning_effort | No | Shift handoff — Select none, low, medium, high, xhigh or max, The default is medium. |
temperature | No | Shift handoff — Temperature applies only at reasoning effort none. |
top_p | No | Shift handoff — Use |
tools | No | Shift handoff — Describe the functions your application can execute and validate their arguments before running them. at most 128 items |
parallel_tool_calls | No | Shift handoff — Independent function requests can arrive together, The application controls their execution. Default: |
tool_choice | No | Shift handoff — Allow automatic selection, forbid calls, require a call or pick one named function. |
response_format | No | Shift handoff — Choose plain text, a JSON object or a JSON Schema response. |
Pricing
Input $1.60 / Output $9.60 per 1 million tokens; Prompts over 272K tokens: Input $3.20 / Output $14.40
Use cases
For a maintenance change, provide the existing behavior and a small before-and-after example. Ask Terra to modify the owning module and explain which caller assumptions might change. Confirm the proposed behavior with the checks associated with that module.
For a weekly operations digest, submit labeled incidents and a fixed reporting window. Request a concise account of customer impact, resolved items and actions still open. Validate dates and incident identifiers before distributing the digest to its intended readers.
Prompt guide & pro tips
Keep the task boundary explicit: name the deliverable, the information that controls it and the conditions that make it acceptable. Include a counterexample when nearby categories are easy to confuse. If the source material has conflicting dates or values, ask Terra to retain the conflict and identify what would resolve it instead of selecting an unsupported answer.
Use a JSON schema for a programmatic handoff and validate required fields after generation. Set an output allowance large enough for reasoning as well as the visible reply; the maximum is 128,000 tokens. Explicit sampling parameters apply only with reasoning none. For repeated jobs, keep stable instructions consistent, limit concurrency in your application and compare the cost per accepted result at the effort settings that meet your accuracy target.
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-5.6-terra/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.

















