GPT-6 models: Astra, Sol and Luna

Choose a GPT-6 model

Use the interactive family comparison to apply the same token counts and capability requirements. The pricing guide explains cache writes, long context and processing modes; the access self-check separates plan and product rules.

Same family, different workload choices

Astra was the first GPT-6 model, released September 3. Sol and Luna arrived September 22. Their shared context capacity does not make them interchangeable for every task. Start with a representative input, define a correct output, and choose the model whose result passes your checks at an acceptable cost.

A code review should identify a reproducible defect and test the repair. A batch extraction should retain source IDs, use allowed labels and leave missing fields unknown. The example library provides concrete versions of both tasks alongside the original Astra document and game exercises.

Product names and exact model IDs

The API IDs are gpt-6-astra, gpt-6-sol and gpt-6-luna. ChatGPT calls Astra in ordinary Chat GPT-6 Pro. Sol and Luna are available in Work and Codex; the September 22 announcement does not put them in ordinary Chat. Free/Go users have a Luna desktop route.

Pro reasoning mode is a request setting. The API still uses the selected model’s exact ID. This directory will add a separate page when OpenAI publishes another GPT-6 model, with its own verified price, limits and use cases.

Plan your first request

Use the token counter to check raw text size and reserve output space. Use the calculator for request and monthly costs. If your workflow repeats a long stable prefix, compare reads and writes in the caching calculator.

Follow GPT-6 news for announcements. Model details and examples keep their own dates, so a new article does not imply every historical test has been repeated.