OpenAI has published new prompting templates for the GPT-6.2 Astra and Sol models
OpenAI has released new prompting templates for the GPT-6.2 Astra and Sol models – by default the models are more concise and are meant to act without unnecessary questions. In addition, in the API, editing the conversation history now discards earlier reasoning steps, which can affect token costs.
OpenAI has published, in the documentation for the GPT-6.2 Astra and Sol models, new prompting templates that it recommends developers use when switching to these models. According to the company, Astra is intended for the most demanding tasks, while Sol can handle more complex work at a lower price. The templates address writing style, model autonomy, and code writing.
According to OpenAI, by default both models respond more concisely and directly than previous versions, so instructions such as "be concise" are no longer necessary. The company recommends prompts that adopt the user's own terminology, use the active voice, and put key points at the beginning. For longer tasks, the model may try to carry shorthand from intermediate reasoning steps into the final answer – the template therefore requires that the final answer be understandable on its own. The company further recommends that when the user's intent is unclear, the model should rather act on its own based on reasonable and reversible assumptions instead of waiting for confirmation, and should only ask questions in the case of a significant change in task scope, a genuine blocker, a necessary user decision, or before an irreversible action.
For code writing, the template instructs the models to repeatedly make use of existing logic in the repository, avoid unnecessary helper functions and abstractions, and write new tests only on request, according to the repository's requirements, or to catch a specific regression. For frontend work, developers are expected to specify images and typography themselves; by default the model chooses realistic images instead of SVGs unless they are requested.
According to the source, the API has also undergone a change: starting with GPT-6.2, editing the conversation history in the Responses API discards the model's earlier reasoning steps, which can affect both token costs and response quality. One of the sources further states that the "temperature" and "top_p" parameters have disappeared from the API. See the source article for details.
Why it matters
Developers switching to the GPT-6.2 Astra or Sol models must update their existing system prompts, otherwise they may encounter unexpected behavior – the models respond more concisely and tend to act directly rather than ask questions. In addition, companies with API integrations need to take into account that editing the conversation history now discards the model's earlier reasoning steps, which affects token costs and can impact response quality in existing deployments.
Two audiences, two different impacts
What this means
For individuals
Developers using the GPT-6.2 Astra or Sol models should adjust the instructions they give, because by default the models respond more concisely and are meant to act independently instead of waiting for confirmation.
For a business
Companies running integrations based on GPT-6.2 Astra or Sol must rewrite their system prompts (otherwise the models respond differently than the companies expect) and check how the new logic for discarding conversation history in the API affects token costs and output quality.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.