OpenAI publishes a guide to the GPT-6 model family
GPT-6 guide helps teams choose Astra, Sol, or Luna for faster workflows Use caching, compaction, and tool use to cut cost and ship smarter at scale
OpenAI has released a practical guide for using the GPT6 model family in product and developer workflows. Published on October 2, 2026, the guide explains how to choose between GPT6 Astra, GPT6.1 Sol, and GPT6 Luna based on the task, cost, and latency needs.
The article recommends using caching and compaction to manage context and reduce costs in production, along with monitoring, data controls, and representative testing before deployment. It also outlines how developers can tune prompts, define skill boundaries, and specify when a model should act independently or ask for input.
The guide further covers longrunning work, including midturn steering, asynchronous tool use, delegation, and computer use for interacting with websites and desktop apps. OpenAI says the models can support workflows ranging from coding and research to structured summaries and other repeated tasks at scale.