GPT-6.1 Sol · Prompt Lab

State the result. Say when you are done.

Compose a Sol prompt and its API settings here. Nothing is sent anywhere.

Guidance & sources ↓

Your brief

What kind of work?

Outcome (required)

Name a concrete result. Add process steps only where they matter.

Context

Done when

Boundaries

Name the actions that need approval, such as deploying or deleting. Everything else can proceed.

Deliverable, length and style

OpenAI's GPT-6 guidance documents no verbosity setting, so say how long and in what style. Blank adds a short-paragraph default.

Reasoning effort

Effort is an API or app setting. The prompt cannot change it, so it appears only in the API settings below.

Agent with tools

Adds a working agreement: finish verification, delegate to subagents when it helps, approval last. Existing approval limits still apply.

Reference material

Paste source text

0 / 16,000 characters. Wrapped and marked as untrusted data. Labels help; they are not access controls.

Builds and copies in your browser. No model call.

Your prompt

Not built yet

A focused starting point

Outcome. Evidence. Boundaries.

Build a prompt from your brief, or start with an example.

Copy into your chosen GPT-6.1 Sol app when ready.

API settings

{
  "model": "gpt-6.1-sol",
  "reasoning": {
    "effort": "medium"
  }
}

Model id gpt-6.1-sol. Efforts: low, medium (default), high, xhigh, max. none is not supported.
Tool calling requires the Responses API.
Omit temperature, top_p and top_logprobs. In Chat Completions also omit logprobs.
To change effort mid-conversation, add a configuration_update input item and leave the request-level value unchanged. Check the compatibility limits first.

Source: model page ↗ and latest-model guide ↗.

Saving Sol prompts is not available yet. Astra, Fable and Opus prompts can be saved.

What the guidance says

Sol follows the general GPT-6 guidance

OpenAI publishes no Sol-specific prompting section. Its prompt guidance is written around GPT-6 Astra and says to evaluate prompts with the model you use. Checked Nov 11, 2026.

Short paragraphs, few lists

Ask for paragraphs that each develop one idea. Use lists only for parallel, sequential, or comparative content.

OpenAI: paragraphs and lists ↗

Say how much to delegate

The guidance suggests delegating in parallel when it saves time or improves quality. State when, if you want it.

OpenAI: subagents ↗

Approval comes last

Finish the required work first so approval is the last step, with a concrete result to review. Reversible and read-only actions need none.

OpenAI: approval ↗

Right-size the tests

Skip extra tests for reversible, low-impact changes. Run what suits the change and widen only when new changes or failures call for it.

OpenAI: testing ↗

Set length and style yourself

OpenAI's GPT-6 guidance documents no verbosity setting. The guide observes that GPT-6 Astra tends toward detailed, formatted answers and says to evaluate its prompts with your own model, so state the structure and style you want.

OpenAI: writing style ↗

Your instructions win

User instructions take precedence over skill guidance, and skill files and AGENTS.md carry weight. Audit them.

OpenAI: skills and files ↗

Checked Nov 11, 2026 against the Sol model page, the latest-model guide and its prompting best practices. This builder composes text only; it does not run or evaluate Sol.