OpenAI publishes GPT-6 family builder guide

OpenAI has released a developer guide for its GPT-6 family. It covers how to choose a model, write instructions, manage long-running work and prepare for production. Most of it is prompting and workflow advice, but it also confirms several product details.

Models. The guide names three models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. It positions Luna for "focused tasks at scale" and repeated work with a clear goal, such as extracting invoice fields, classifying requests or producing structured summaries. The excerpt does not describe how Astra and Sol differ in workload, pricing or benchmarks.

Speed and reasoning. In the API, Fast mode gives faster, more consistent response times at a higher per-token cost than Standard processing. In Codex, OpenAI suggests starting with each model's default reasoning level and lowering or raising it depending on the task.

Long-running work. OpenAI says the GPT-6 family can take on tasks "that span hours or days." In the API:

• Asynchronous tool calling lets the model keep doing independent work while a slower tool, such as a test suite, runs.
• GPT-6.1 Sol supports multi-agent workflows in the Responses API. It can hand independent subtasks to subagents and combine their findings into one response. The feature is in beta.

In Codex, GPT-6 Astra can ask for clarification mid-task, and users can steer an active task when requirements change.

Computer use. All three models support computer use, so they can work with websites and desktop apps that have no API. OpenAI recommends using an API or connected tool when one can do the job directly, and saving computer use for steps that need reading a screen, clicking or filling in forms.

The guide also advises keeping skills descriptions short, updating AGENTS.md files and setting explicit approval boundaries instead of blanket "always ask" rules.