OpenAI.

In plain English

OpenAI's model for the hardest generally available work, and its most expensive.

Good for

Difficult research, coding, computer use, and documents where a weaker model would change the result.

Things to be aware of

Using Astra to pull out simple information or write routine summaries is a poor use of money. Luna does that for about a hundredth of the price.

In plain English

OpenAI's main business model, with strong performance at a fifth of Astra's price.

Good for

Coding, multi-step professional work, computer use and agents.

Things to be aware of

Try it on your own real work before paying for Astra. Astra costs five times as much, and Sol may well do the job.

In plain English

A faster, cheaper OpenAI model for visual and on-screen work, at a tenth of Astra's price.

Good for

Reading invoices, technical drawings and software screens, and drafting everyday business text.

Things to be aware of

Demos use neat, tidy examples. Before you rely on it, test it on your own messy scans and real systems.

In plain English

OpenAI's cheapest GPT-6 model, at about a hundredth of Astra's price, built for work at scale.

Good for

Sorting, pulling out information, routing and short summaries.

Things to be aware of

Cheap models still make mistakes. Decide what a person checks, and when work moves up to a stronger model.

In plain English

OpenAI's previous workhorse, still widely used in existing systems.

Good for

Workflows that have already been tested on this model.

Things to be aware of

No retirement date has been announced, and OpenAI points people to it as the replacement for many older models it's retiring. It was one of the two models in the July 2026 Hugging Face incident.