OpenAI has published a new set of system prompt templates and API guidelines for its latest models, GPT-6.2 Astra and Sol, designed to tackle issues with overly brief answers, unwanted follow-up questions, and clunky code generation.
The company advises developers migrating to the new models to update their system prompts rather than relying on legacy configurations. The updates target specific behaviors observed in Astra—built for high-demand tasks—and Sol, which is engineered for cost-efficiency.
Refining Writing Style and Autonomy
Because Astra is naturally more concise than its predecessors, OpenAI recommends rethinking standard instructions like "be concise." For general productivity tasks, developers should encourage the model to adopt active voice, prioritize key points, and use user-supplied terminology. To prevent the model from leaking internal reasoning shorthand into final outputs, OpenAI introduced explicit instructions for self-contained answers:
Treat your final answer as the user's first look at your work, and not as a continuation of intermediate commentary. The final answer must always be fully self-contained: users should never need to read earlier commentary updates for context...
Additionally, OpenAI is addressing the model's tendency to pause and ask for confirmation when intent is ambiguous. The updated guidelines instruct GPT-6.2 to act on reversible assumptions and push forward with authorized next steps, stopping only when facing significant scope changes, true blockers, or destructive actions.
Streamlining Code Generation
For software development, GPT-6.2's new templates push the model to write cleaner, more efficient code by reusing existing repository logic and avoiding unnecessary helper functions or abstractions. The guidelines dictate that new tests should only be added when explicitly requested, required by the repository, or necessary to catch specific regressions. Front-end development tasks should similarly avoid redundant filler badges, decorative numbering, and unnecessary helpers.
API Adjustments and Reasoning Content
Alongside prompt updates, OpenAI introduced changes to the Responses API. Starting with GPT-6.2, editing conversation history or switching certain parameters can cause the API to discard earlier reasoning content, potentially increasing token usage and repetition. For new integrations, OpenAI recommends a "medium" reasoning level, while existing setups should transition from "none" or "minimal" to "low."
Launched in mid-November, Astra's API pricing begins at $8 per million input tokens and $40 per million output tokens for contexts up to 272,000 tokens, after which pricing scales higher and Codex automatically compresses context.