Introduction
GPT 6.1 Sol reviews difficult code changes and professional analyses with text and image evidence in one conversation. Keep code, failing behavior and acceptance criteria together when asking Sol for a repair; ask it to identify which file supports each finding.
| Field | Value |
|---|---|
| Input | Text |
| Output | Text |
| Context window | 1,050,000 tokens |
| max_tokens ceiling | 128,000 |
| Inputs | text / images |
Key features & highlights
The GPT 6.1 model you reach for when the task is difficult
GPT 6.1 Sol reviews difficult code changes and professional analyses with text and image evidence in one conversation. Keep code, failing behavior and acceptance criteria together when asking Sol for a repair; ask it to identify which file supports each finding.
Five effort levels, from quick to exhaustive
reasoning_effort accepts low, medium, high, xhigh and max, and medium is the default. Sol has no none setting, so even the lightest request gets a short reasoning pass. Start at medium, step up for problems where a missed detail is expensive, and keep max for the cases that justify the extra tokens.
Review a repository with its specification beside it
The 1,050,000-token window lets you keep a repository, a specification and the question together instead of summarising. Prompts longer than 272,000 tokens fall into a separate long-prompt pricing tier, so it pays to send the whole codebase only when the task needs it.
Function calls and screenshots in the same conversation
Sol can propose functions through this platform’s compatibility interface, with up to 128 definitions; your application executes them and returns the matching tool_call_id. OpenAI’s native tool interface uses Responses.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Include Sol’s code, error trace and review criteria in messages; keep tool results beside the calls they answer. at least 1 items · at most 256 items |
max_tokens | No | Reserve |
reasoning_effort | No | Sol offers low, medium, high, xhigh and max; begin a difficult code review at medium. |
stop | No | For Sol, supply one stop string or up to four; choose markers that will not truncate a patch. at least 1 characters · at least 1 items · at most 4 items |
temperature | No | Adjust Sol’s temperature only if your code-review evaluation calls for different sampling. |
top_p | No | Use |
tools | No | Define the functions Sol may request for inspecting code; your application validates and runs each function. at most 128 items |
parallel_tool_calls | No | Enable independent Sol inspection calls in the same turn when their results do not depend on one another. Default: |
tool_choice | No | Use auto, none, required or a named function to control Sol’s next tool step. |
response_format | No | Specify the format of Sol’s findings, then validate any returned JSON before using the proposed change. |
Pricing
Input $2 / Output $10 per 1 million tokens; Prompts over 272K tokens: Input $4 / Output $15
Use cases
Review a change against the whole repository
Keep code, failing behavior and acceptance criteria together when asking Sol for a repair; ask it to identify which file supports each finding.
Work through a multi-step analysis
Put the source tables, the policy text and the question in one request and ask for findings with their evidence. Ask for a fixed output shape, such as finding, source and confirming check, and validate it in your application.
Run an agent that can act
Sol can propose functions through this platform’s compatibility interface, with up to 128 definitions; your application executes them and returns the matching tool_call_id. OpenAI’s native tool interface uses Responses.
Return tool results with the matching id
Specify the format of Sol’s findings, then validate any returned JSON before using the proposed change.
Prompt guide & pro tips
Start at medium and move one step
Sol offers low, medium, high, xhigh and max; begin a difficult code review at medium.
Give
max_tokensroom for reasoningmax_tokensdefaults to 4,096 and accepts up to 128,000, and the model's reasoning is counted inside it. For a long patch or a detailed report, set it well above the length you expect to read, particularly at xhigh and max, and checkfinish_reasonbefore treating a reply as complete.Put the material first and the question last
Place documents or code early, label each source, and ask the question at the end. Keep instructions in the system message so quoted text is not mistaken for your request. Watch the 272,000-token mark if cost matters.
Return tool results with the matching id
When the model calls a function, send your result as a tool message carrying the same
tool_call_id, and keep the assistant'stool_callsin the history. For machine-readable output,response_formatand tool parameters both accept a JSON schema; validate whatever comes back.
GPT 6.1 Sol
GPT 6.1 Sol reviews difficult code changes and professional analyses with text and image evidence in one conversation.
openai/gpt-6.1-sol/chat
Keep code, failing behavior and acceptance criteria together when asking Sol for a repair; ask it to identify which file supports each finding.
Sol’s 128,000-token reply budget includes reasoning as well as the patch or analysis you see; the default cap is 4,096.
Sol can propose functions through this platform’s compatibility interface, with up to 128 definitions; your application executes them and returns the matching tool_call_id. OpenAI’s native tool interface uses Responses.
For a Sol repository review, input above 272,000 tokens moves the entire request to the long-prompt tier; consult this page’s current rates before sending the full archive.
GPT 6.1 Sol is intended here for code and professional work; NSFW availability is not guaranteed, and the model’s content rules still apply.
GPT 6.1 Sol is intended here for code and professional work; NSFW availability is not guaranteed, and the model’s content rules still apply.
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-6.1-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.