Introduction
GPT 5.6 Sol is the flagship tier of the GPT 5.6 family for complex professional work. It accepts text and images and produces text, with room for 1,050,000 tokens across the working context. The input ceiling is 922,000 tokens, and a response may use up to 128,000 tokens for its reasoning and final deliverable.
Key features & highlights
The reasoning range includes none, low, medium, high, xhigh and max. Medium is the starting setting for an ordinary reasoning request. Allocate a larger output allowance when the task requires exploring several alternatives; choosing max with a tiny cap can exhaust that allowance before the model writes the report you expected.
Structured replies help turn a long analysis into fields a downstream reviewer can inspect. Images can contribute a chart, a diagram or a scanned page to that evidence. Function definitions connect the discussion to application-controlled data retrieval and actions, while argument validation and permission checks remain with the application that executes each operation.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Describe the audience and the decision your report must support. Give the required section order and define how uncertainty should be represented. at least 1 items · at most 256 items |
max_tokens | No | Design review — The output limit covers reasoning and visible text together. |
reasoning_effort | No | Design review — Select none, low, medium, high, xhigh or max, The default is medium. |
temperature | No | Design review — Temperature applies only at reasoning effort none. |
top_p | No | Design review — Use |
tools | No | Design review — Describe the functions your application can execute and validate their arguments before running them. at most 128 items |
parallel_tool_calls | No | Design review — Independent function requests can arrive together, The application controls their execution. Default: |
tool_choice | No | Design review — Allow automatic selection, forbid calls, require a call or pick one named function. |
response_format | No | Design review — Choose plain text, a JSON object or a JSON Schema response. |
Pricing
Input $4 / Output $24 per 1 million tokens; Prompts over 272K tokens: Input $8 / Output $36
Use cases
For a specification review, label each version and ask Sol to identify requirements that changed meaning. Require a source label and a proposed acceptance criterion for every reported conflict. Have the reviewer distinguish a true contradiction from wording that merely needs clarification.
For a research synthesis, separate primary documents from commentary before sending them. Request an evidence table, the assumptions used to combine sources and the questions left unresolved. Retain the supplied material so the resulting citations can be checked.
Prompt guide & pro tips
Describe the audience and the decision your report must support. Give the required section order and define how uncertainty should be represented. If an example answer is included, label its role clearly so its invented values do not become evidence. For code work, ask for compatibility checks tied to the interfaces that callers actually use.
Keep sampling changes on reasoning none; explicit temperature or top_p with enabled reasoning is rejected. Track the measured input, cached input, cache writes and output rather than treating the requested output cap as the consumed quantity. Long prompts above 272,000 input tokens use a different price tier for the full request. Review both the reported completion state and the substantive findings before passing a result to the next workflow step.
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-sol/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.

















