Brief IA
›

OpenAI Recommends Revisiting Brevity Guidelines

Code & Dev·Tom Levy·

OpenAI Recommends Revisiting Brevity Guidelines

OpenAI Recommends Revisiting Brevity Guidelines
Key Takeaways
1GPT-6.2 Astra and GPT-6.2 Sol were launched on November 17.
2OpenAI recommends reviewing brevity guidelines when migrating to Astra.
3The API will drop reasoning elements if it detects certain changes in the history, OpenAI announces.
Why it matters : According to OpenAI, losing these elements can degrade responses and increase token consumption to redo previous work.

OpenAI recommends developers to review their brevity guidelines for GPT-6.2 Astra, launched on November 17 alongside GPT-6.2 Sol. The company describes Astra as more direct by default than its predecessors. The migration also includes a change announced in the retention of reasoning by the API.

Modifying history may erase reasoning

OpenAI announces that starting with GPT-6.2, a modification of the history by the client detected by the API will result in the loss of reasoning elements. The same treatment is expected after switching to a lower version model or changing the account or organization of the end user. These elements contain reasoning derived from previous exchanges. According to the company, their removal can degrade responses and require additional tokens to redo work that has already been done.

Examples provided by OpenAI include a shortened tool output, old exchanges replaced by a summary, and a rewritten developer message. The latter carries the application's instructions. To evolve them, the company recommends adding a new developer message that requests the application of new guidelines while ignoring the previous ones, without altering the original. Some writing tools may modify their instructions during the conversation: this practice could therefore lead them to lose previous reasoning and incur additional token expenditure.

Astra can await a human decision

When the request lacks precision, GPT-6.2 Astra may propose the continuation of the work and then wait for agreement, OpenAI indicates. For tasks to be carried out autonomously, the company advises explicitly defining the expected outcome and the decisions that belong to the user. The proposed instructions allow for reasonable, reversible, and explicit assumptions to continue the work.

A halt should then occur if the scope changes significantly, if an obstacle blocks execution, or if a human decision is necessary. A check with the user is also planned before any destructive or irreversible operation. OpenAI also describes an asynchronous tool: the model can ask a question without interrupting work that does not depend on it. This tool must be defined by the application, which displays the question; it is not provided directly in the response API.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Tailoring texts to the reader and the task

For emails, documents, or other writings addressed to an external audience, OpenAI recommends specifying the readership and providing contextual elements that they may not know. In revision, the proposed guideline is to retain the original content and formulations, subject to changes requested by the user. The company also describes an adjustable tone, ranging from professional colleague to conversational assistant. Its example intended for daily work asks to adopt the user's vocabulary and sources, then to place the essentials at the forefront.

Another precaution concerns responses produced after lengthy tasks. According to OpenAI, they may retain abbreviations, jargon, or shortcuts from intermediate reasoning or other agents. The company recommends requesting a final standalone text, understandable by someone who has not followed the progress reports, without these shortcuts.

Settings to review during migration

Writing tools transitioning from GPT-6 Astra to the new version are particularly affected by the review of instructions. OpenAI described the old model as willingly detailed, with lists, tables, and Markdown intended to facilitate reading. For less formatted prose, the company suggested asking for concise paragraphs and reserving lists for parallel or sequential data. It had already noted that GPT-5.6 was generally less verbose than GPT-5.5 and that some tasks no longer required an additional brevity guideline.

Astra and Sol are accessible via the API identifiers gpt-6.2-astra and gpt-6.2-sol. OpenAI presents its examples as a starting point based on observed behaviors on these models. The GPT-6 recommendations remain accessible in a tab on the same page as those for GPT-6.2, to compare guidelines between versions.

For new integration, OpenAI advises a medium reasoning effort. Applications previously set to zero or minimal effort are encouraged to choose the low level.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.