Concrete Draft Before Clarifying Questions
Generate a complete working draft before asking clarifying questions, ensuring assumptions are clearly marked.
GPT-6 Sol
LLMGPT-6 Sol is the reasoning tier of OpenAI's GPT-6 family below Astra: strong multi-step reasoning for demanding work, priced 50% below its GPT-5.6 predecessor. It shares Astra's bias-toward-action behaviour and instruction following, supports the none reasoning effort for fast turns, and pairs naturally with Luna as the tier that hard cases escalate to.
How to use GPT-6 Sol
Set model to gpt-6-sol. Pick reasoning_effort per call: none for fast, well-specified work, low or medium for most tasks, high for hard analysis. Use the Responses API when you need reasoning plus tools or strict structured output. Keep your Astra prompts; they carry over without change.
Best for
Demanding reasoning at scale where Astra is more than the job needs: analysis, planning, code review, hard-case escalation behind a faster model, and agent leads that delegate to subagents.
How it behaves
Follows instructions closely, biases toward completing the task rather than asking, and reasons at length when effort is raised. Supports reasoning_effort none for latency-sensitive turns. Function calling on Chat Completions works only at none; reasoning with tools needs the Responses API. With reasoning on, temperature and top_p are ignored and should be removed.
Tips
State the outcome, the autonomy policy and the stop conditions in the system prompt. Ask it to produce a concrete draft before clarifying questions. Use it as the escalation tier for cases a faster model marked uncertain. Remove temperature and top_p whenever reasoning is on.
Pricing
Available in ChatGPT paid plans and via the OpenAI API. API prices are 50% below GPT-5.6 promotional pricing; see the OpenAI pricing page for current per-token rates.
Prompts for GPT-6 Sol
See all 6Bias Toward Action Operator Preamble
Generate an action-oriented assistant prompt focused on task completion and minimal clarification requests, outputting a structured report of actions and assumptions.
User Instruction Precedence Over Loaded Skills
Defines instruction precedence for a workspace assistant in interactions within a specified workspace.
Subagent Parallelisation Instruction
Facilitates the delegation of independent tasks using collaboration tools while maintaining oversight on the main project goal, designed for efficient project management output.
Plain Paragraph Output Policy
Generate clear and concise response templates through structured writing rules, ensuring clarity in discussions.
Hard Case Handoff to the Reasoning Tier
Analyze and resolve cases where a fast-tier AI model was uncertain, following decision rules.