Introduction
GPT 6 Sol is built for coding and agent workflows that move through several dependent steps. A conversation can contain source files, test output and images of the interface under investigation. Its 1,050,000-token context can accommodate substantial supporting material, while the generated response has a 128,000-token ceiling shared by reasoning and visible text.
Key features & highlights
Six reasoning settings, from none through max, let an application allocate thought to the current step. A request to copy a known identifier and an investigation of a failing migration need different budgets. The default is medium; raising effort can consume more of the same output allowance before a visible answer appears.
Function calls provide a way for Sol to request operations implemented by your application. Define narrow arguments for reading a file, running a check or retrieving a record, then decide which calls may execute. The model does not acquire a terminal merely because a function is named run_command. This endpoint uses the Responses transport for the conversation and its tool exchanges.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Start a coding conversation with the desired behavior, the files that own it and a concrete way to verify the change. Identify generated files and external text as evidence. at least 1 items · at most 256 items |
max_tokens | No | Regression trace — The output limit covers reasoning and visible text together. |
reasoning_effort | No | Regression trace — Select none, low, medium, high, xhigh or max, The default is medium. |
temperature | No | Regression trace — Temperature applies only at reasoning effort none. |
top_p | No | Regression trace — Use |
tools | No | Regression trace — Describe the functions your application can execute and validate their arguments before running them. at most 128 items |
parallel_tool_calls | No | Regression trace — Independent function requests can arrive together, The application controls their execution. Default: |
tool_choice | No | Regression trace — Allow automatic selection, forbid calls, require a call or pick one named function. |
response_format | No | Regression trace — Choose plain text, a JSON object or a JSON Schema response. |
Pricing
Input $1.60 / Output $8 per 1 million tokens; Prompts over 272K tokens: Input $3.20 / Output $12
Use cases
For a regression investigation, provide the failing assertion, a small reproducer and the relevant implementation. Ask for a proposed cause, the smallest discriminating experiment and a patch only after the evidence supports that cause. Return the experiment result in the next turn.
For an interface repair, attach a screenshot and describe the intended interaction. Include the component contract and the viewport that matters. Request a change tied to a visible defect, then inspect the rendered result alongside the acceptance checks.
Prompt guide & pro tips
Start a coding conversation with the desired behavior, the files that own it and a concrete way to verify the change. Identify generated files and external text as evidence. When a tool returns an error, preserve its relevant details so the next turn can distinguish an implementation defect from an unavailable dependency.
Budget for the whole answer, including reasoning, and review the terminal status before treating an empty result as success. Use none when explicit temperature or top_p sampling is needed; reasoning turns use default sampling. Keep unrelated repository history out of a narrow fix. A large context is useful when those documents affect the decision, and requests above 272,000 input tokens enter the long-context price tier.
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-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.

















