OpenAI publishes a GPT-6 family model-selection guide

The short version

OpenAI published a guide for choosing among GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. It connects task difficulty with model choice and recommends evaluating task success, latency and cost in production workflows.

What changed

The new publication explains model selection and workflow design; it is not a new model-release announcement.

What it means for you

Developers can test routing choices against their own tasks instead of assigning the largest model to every request.

The guide positions Astra for the hardest reasoning, GPT-6.1 Sol for coding, research and computer use, and Luna for focused, repeated tasks. These are the vendor’s recommended roles rather than a universal ranking for every workload.

It discusses consistent prompts, reusable skills, repository instructions, prompt caching and compaction. For longer work it also covers asynchronous tools, steering and delegation. A practical follow-up is to compare completion quality, latency and cost on the same representative tasks before changing default routing.

The event is publication of a guide. It does not establish October 2 as the release date of the model family. The exact timestamp is taken from OpenAI’s official RSS entry, whose article displays an October 2 date.

Fact check

VerifiedThe guide positions Astra for hardest reasoning, Sol for coding/research/computer use and Luna for focused repeated work.Evidence

VerifiedThe guide recommends workflow evaluations of success, latency and cost and discusses prompts, skills, repository instructions, caching, compaction, asynchronous tools, steering and delegation.Evidence