Intelligence Report

OpenAI updates prompt guidance and API behavior for GPT-6.2 Astra and Sol

New templates target terse replies, needless questions and excess test writing, while the Responses API now drops prior reasoning in some cases.

INSIGHTEX Editorial Desk

2 min read

OpenAI updates prompt guidance and API behavior for GPT-6.2 Astra and SolIntelligence Report

What happened

OpenAI has published new system prompt templates for its GPT-6.2 Astra and Sol models, meant to correct habits such as terse replies, needless clarifying questions and over-eager test writing in coding tasks, The Decoder reported. The company told developers moving to either model to revisit their existing system prompts. Astra targets the hardest workloads, while Sol is positioned for complex jobs at a lower price.

OpenAI says Astra now answers more tersely by default, so prompts demanding brevity may need rethinking. The templates tell the models to carry out authorized next steps themselves, note any reversible assumptions, and pause only for big scope changes, real blockers, user decisions or destructive actions. For coding, they favor reusing existing logic, readable names, and adding tests only when asked, mandated by the repository or aimed at a particular regression.

On the API side, starting with GPT-6.2 the Responses API drops prior reasoning content when developers edit conversation history, move to certain other models, or change the end user or organization. OpenAI says this can lower answer quality and raise token use as work gets repeated. It recommends the medium reasoning setting for new integrations, suggests moving setups on none or minimal to low, and no longer supports temperature or top_p.

Context

In April, OpenAI urged developers to abandon old prompt stacks for GPT-5.5 in favor of short, outcome-focused prompts. In September it issued GPT-6 Astra prompts to curb follow-up questions, and OpenAI developer Eric Provencher warned that bloated skills and AGENTS.md files slow the model. GPT-6.2 Astra and Sol launched on November 17, with Astra priced at $8 per million input and $40 per million output tokens up to 272,000 tokens.

Astra profile

SourceThe Decoder

The briefing

The week in tech, in one email.

The stories worth your time, every week. Unsubscribe anytime.