OpenAI publishes GPT-6.2 Astra and Sol prompt templates and API changes

OpenAI has published new prompt templates in the documentation for GPT-6.2 Astra and Sol, according to The Decoder. They cover writing style, initiative and coding. OpenAI says developers moving to either model should review their system prompts. The templates are meant to curb behaviors the company observed in both releases. Astra is built for the most demanding tasks, while Sol handles complex work at lower cost. The Decoder reports that OpenAI launched both models on November 17.
Writing style. Astra writes more briefly and directly by default than earlier models, so developers switching over should reconsider instructions like "be concise." For use as a colleague in everyday work, OpenAI recommends a prompt that adopts user and source terminology, uses active voice and puts key points first. When writing for external readers, the model should account for their lack of context. When editing, it should preserve the author's meaning and phrasing. On long tasks the model sometimes carries shorthand from its reasoning, or from other agents' messages, into the final response. A dedicated prompt tells it to write the final answer so it stands on its own, as the user's first look at the work, with no need to read earlier commentary and no working jargon or abbreviations carried over from its reasoning, tool calls or instructions to subagents.
Initiative. OpenAI says that when user intent is unclear, GPT-6.2 sometimes suggests a next step and waits for confirmation. The recommended prompt tells the model to carry out those steps itself whenever it is authorized to do so. It should pause only for a major change in task scope, a real blocker, a decision the user must make, or an irreversible or destructive action. When details are unclear, the model should make reversible assumptions, flag them and keep working. OpenAI adds that higher reasoning effort also helps the model follow through on final checks during long tasks.
Coding. OpenAI says GPT-6.2 writes more efficient code than its predecessors. The template tells it to reuse existing classes, methods, representations and APIs in the user's repo, and to avoid unnecessary helper functions and abstractions. New tests should be added only when requested, required by the repo, or needed to catch a specific regression that existing tests would miss. Fewer lines and shorter variable names are not a substitute for simplicity, and code-golfed expressions that hide branches or assumptions count as overly complex. For frontends, developers should specify imagery and typography directly. By default the template favors realistic images from image generation or search and uses SVGs only on request. Decorative numbering, filler badges and helper text that only repeats what the interface already shows should be left out.
API changes. Starting with GPT-6.2, the Responses API discards earlier reasoning content if developers edit conversation history, switch to certain other models, or change the end user or organization. That can hurt response quality and increase token usage, because the model repeats work. For new integrations OpenAI recommends the "medium" reasoning level, while existing setups using "none" or "minimal" should move to "low." Parameters like temperature and top_p aren't supported anymore.
Background and pricing. In April, OpenAI advised against carrying over old prompt stacks for GPT-5.5 and urged developers to start fresh with short, outcome-focused prompts. September brought GPT-6 Astra prompts to curb frequent follow-up questions and ban AI cliches. OpenAI developer Eric Provencher warned then that bloated skills and AGENTS.md files slow the model down. Back then even small changes triggered thorough testing from Astra; GPT-6.2 is now meant to write new tests only on request. Astra's API pricing starts at $8 per million input tokens and $40 per million output tokens for up to 272,000 tokens. Beyond that limit Astra 6.2 costs $16 and $60, with cache reads at $0.80 per million tokens. Astra 6.0 costs more beyond the cutoff: $20 for input and $75 for output per million tokens. Because longer coding or work sessions can pass that cap quickly, Codex automatically compresses context at 272,000 tokens. Picking the right model comes down to budget and quality needs, according to OpenAI's Ted Sanders.
Key facts
- OpenAI published copy-ready prompt templates for GPT-6.2 Astra and Sol on writing style, initiative and coding, and says developers moving to either model should review their system prompts.
- Astra writes more briefly by default; the initiative template tells the model to act on reversible assumptions and pause only for a scope change, a real blocker, a user decision, or an irreversible or destructive action.
- The coding template tells the model to reuse existing logic, avoid needless helpers, and write new tests only when requested, required by the repo, or needed for a specific regression.
- From GPT-6.2 the Responses API discards earlier reasoning content when conversation history is edited, certain other models are switched in, or the end user or organization changes; temperature and top_p are no longer supported.
- Astra API pricing starts at $8 per million input and $40 per million output tokens up to 272,000 tokens, then $16 and $60 for Astra 6.2 beyond that.
Why it matters
Prompts written for older models can work against GPT-6.2. Astra already answers more briefly by default, so a habitual "be concise" may now over-trim. The model also sometimes stops to suggest a next step instead of taking it, which interrupts agentic workflows. OpenAI's templates target exactly these behaviors. The API changes matter just as much: reasoning content is now dropped in several situations, which can degrade answers and raise token use. This follows earlier guidance from OpenAI, including the April advice to start fresh for GPT-5.5 and the September GPT-6 Astra prompts against frequent follow-up questions and AI cliches.
Who it affects
Developers moving existing products to GPT-6.2 Astra or Sol, especially those with long system prompts, agent loops, coding assistants or frontend generators. It also touches anyone who edits conversation history, switches models mid-conversation, or changes the end user or organization in the Responses API, and anyone still setting temperature or top_p. Teams running Codex sessions are affected by the 272,000-token point, where Codex automatically compresses context.
How to use it
OpenAI's guidance, as reported: review existing system prompts when switching, and reconsider instructions like "be concise." Use the published templates from the models' documentation for writing style, initiative and coding. Tell the model to write final answers that stand on their own, and to carry out authorized next steps itself. For frontends, specify imagery and typography directly. For reasoning effort, use "medium" for new integrations, and move setups using "none" or "minimal" to "low." Remove temperature and top_p. Avoid editing conversation history where possible, since that discards earlier reasoning. On cost, Astra starts at $8 per million input tokens and $40 per million output tokens for up to 272,000 tokens, and Astra 6.2 costs $16 and $60 beyond that. The article gives no price for Sol.
How solid is it
The substance is OpenAI's own documentation, relayed by The Decoder, which reproduces template passages and attributes the behavioral observations and recommendations to OpenAI. Some statements, such as Astra being built for the most demanding tasks and the pricing, are in The Decoder's own voice. No benchmark or measured results are given for the prompt templates, so the effect of the templates on output quality is not demonstrated in the article.
Risks and caveats
The article does not state which "certain other models" trigger the discarding of reasoning content, so developers who switch models mid-conversation have to check the documentation. It does not say whether the 6.2 templates replace or only supplement the September GPT-6 Astra prompts. No years are given for November 17, April or September. The template tells the model to act on assumptions and to flag them, and it should still stop before irreversible or destructive actions, so developers should check what authorization their setup actually grants. Removing temperature and top_p may break existing integrations that set them.
“Only stop to clarify with the user if there is significant change in task scope, you are concretely blocked, have a decision that the user needs to resolve, or before irreversible or destructive actions.”
— OpenAI's GPT-6.2 prompt template, as reproduced by The Decoder