OpenAI is suggesting that developers who are migrating to GPT-6.2 Astra should revisit instructions specified as “be concise” whenever they update their scheme prompts. The guide for developers on how to use GPT-6.2 states that Astra produces output that is additional straightforward and concise by default than previous GPT models.
The company’s guidance for GPT-6 Astra says the example tends toward detailed, formatted responses.
GPT-6.2 Astra and GPT-6.2 Sol launched Nov. 17. Developers can use the two in the API as gpt-6.2-astra and gpt-6.2-sol.
How The Guide Describes GPT-6.2 Astra’s Writing
Astra is described in the GPT-6.2 guide as “more concise and straightforward by default than before models,” alongside an adjustable general spirit that can range from a expert coworker to a additional conversational assistant.
The GPT-6.2 guide provides an example immediate for apps that use the example as a collaborator in day-to-day work. This immediate tells Astra to rod alongside the terminology of the person and its sources and to location the key points at the outset of its response.
The GPT-6 direction states, “GPT-6 Astra tends to use lists, tables and Markdown to create responses scannable.” The GPT-6 guide additionally contains an example immediate for apps that necessitate prose alongside small formatting. This example immediate requests “clear, concise paragraphs” and directs the example to reserve lists for presenting parallel or sequential data.
The GPT-5.6 guide states that GPT-5.6 is mostly additional concise by default than GPT-5.5, and that brevity instructions (such as “Be concise” and “Keep it short”) may be unnecessary for several types of task.
Prompts For Drafting, Editing And Final Answers
The GPT-6.2 guide advises asking the example to modify for its mark spectators in emails, documents and any another penning intended for an external reader. It does this by suggesting a immediate which states that an external spectators would likely have small understanding of the user’s context, so the example volition need to provision the backdrop data that spectators needs.
The immediate additionally provides direction on editing by instructing the example to “preserve the author’s satisfied and wording” whenever reviewing or editing existing activity and adhere to the user’s direction as to what changes are to be made.
When performing longer tasks, the example can keep abbreviations, terminology or shorthand from the reasoning process, or from another agents, in its final answers. The guide suggests a method to location this by having the example compose a self-contained reply for anyone who hasn’t been tracking advancement updates. Additionally, the guide’s example immediate tells the example not to contain specified shorthand, terminology or abbreviations in the response.
When The Model Pauses On Long Tasks
The guide says GPT-6.2 Astra may propose a next stage and delay for confirmation whenever a user’s intent is unclear. For activity that can run on its own, the guide’s direction is to enchantment out the outcome the example should complete and which decisions need the user’s input.
The guide’s suggested prompts inform the example to create reasonable, reversible assumptions and keep working, stating what it assumed. It should halt and ask lone whenever the task’s range changes in a big way, whenever it’s blocked, or whenever the person has to create a decision. It should additionally inspect in before irreversible or destructive actions.
The guide additionally points to an async inquiry tool, which lets GPT-6.2 ask the person a inquiry during it keeps operating on item that doesn’t depend on the answer. OpenAI’s async tool calling page says the app defines that tool and shows the inquiry to the user, and that it isn’t a built-in tool in the Responses API.
An API Change For Apps That Edit Earlier Turns
Starting alongside GPT-6.2, the API volition autumn reasoning items as shortly as it identifies that the client has made an edit to previous history, dropped downward to a lesser example version, or changed the end-user document or organization, according to OpenAI’s API documentation. Reasoning items provision the model’s reasoning from before in the conversation. Dropping these can guide to a decrease in answer norm and addition the figure of tokens used by the example to reiterate former work.
Examples on OpenAI’s page concerning the change contain shortening a tool result, rewriting a developer communication (which carries an app’s instructions), and replacing older turns alongside a summary. For new instructions, the leaf suggests appending a developer communication that asks the example to disregard older instructions and prosecute the new ones, in location of editing the original.
Why This Matters
Teams construction example or editing tools on GPT-6 Astra have prompts to assessment previous to migrating to GPT-6.2 Astra. Developers need to revisit any education that asks GPT-6 Astra to create briefer answers, according to the 6.2 guide. Some of these tools may rewrite their own instructions as fine during a conversation. This could guide to GPT-6.2 Astra losing its before reasoning, and spending additional tokens to recreate this process.
Looking Ahead
The GPT-6.2 guide calls its prompts a starting point, according to behavior seen alongside GPT-6.2 Astra and Sol. The GPT-6 direction is motionless on the identical leaf as the 6.2 guide, under its own tab, so teams can difference what all guide says concerning the example they’re moving from.
For reasoning effort, the 6.2 immigration notes propose average for new integrations and low for apps that are currently set to none or minimal. Anthropic’s Claude Opus 5.5 guide, which I covered in September, additionally asked developers to retest attempt settings carried complete from Opus 5.
Featured Image: Tetiana Yurchenko/Shutterstock