Which ChatGPT Work Model Should You Use? Astra, Sol, Terra, and Luna Explained

Retro control console with Astra, Sol, Terra and Luna panels and a model selection dial

The most capable AI model is not automatically the best choice for every job. A short customer follow-up, a spreadsheet cleanup, and a complicated vendor decision need different amounts of reasoning—and different amounts of your time.

OpenAI’s official model guide offers a useful starting point for choosing among Astra, GPT-6.1 Sol, GPT-6.1 Terra, and Luna. The practical business lesson is not to memorize a leaderboard. It is to match the model and its settings to a clearly defined result.

Choose the lightest model and reasoning setting that reliably meets your quality standard—not simply the fastest response or the biggest model name.

First, know which part of ChatGPT this guide covers

The linked documentation describes models for ChatGPT Work and Codex. It should not be read as a promise that every model appears in the ordinary Chat model picker. Available models, tools, controls, and usage limits depend on your plan, client, workspace settings, and rollout.

In the desktop app and ChatGPT Work on the web, OpenAI describes a Power control with faster and smarter options, plus Advanced controls for selecting an available model and reasoning effort. Treat the documentation’s interface examples as illustrations, not a guarantee that your account will show an identical menu.

Four models, four useful starting points

OpenAI positions the models around different workloads. The business examples below are Loudernet’s suggested applications of that guidance—not results from a benchmark we have conducted.

Model OpenAI’s positioning A business task to try
GPT-6 Astra Demanding workflows requiring sustained reasoning and judgment Compare competing website proposals against budget, requirements, and delivery risks
GPT-6.1 Sol Complex, repeated or long-running work where cost matters Turn a project brief, meeting notes, and timeline into a coordinated implementation plan
GPT-6.1 Terra Clear writing, visual understanding, and computer use Review campaign screenshots and draft a concise explanation of the visible changes
GPT-6 Luna Focused, repeatable, high-volume tasks Convert consistently formatted meeting notes into action items with owners and deadlines

These are starting points, not rigid boundaries. A seemingly simple task can become difficult when the source material conflicts, required details are missing, or mistakes would be expensive to repair.

Use Astra when the hard part is the decision

Consider a restaurant choosing between two reservation platforms. One has a lower monthly fee; the other integrates with an existing workflow but brings a longer contract and uncertain migration effort.

The useful deliverable is not a polished summary of both sales pages. It is a requirements-based comparison, a list of unresolved questions, and a recommendation that makes its assumptions visible. That is the sort of demanding, multi-source work for which OpenAI positions Astra.

Give the model the actual proposals, your must-have requirements, and the decision criteria. Ask it to distinguish documented facts from assumptions. Keep final purchasing decisions with the people responsible for the business.

Try Sol when complex work becomes a repeatable process

A marketing agency might prepare a weekly client update from performance notes, campaign changes, and an implementation backlog. The task involves several inputs, but the deliverable follows a known structure.

GPT-6.1 Sol is worth comparing with Astra for that recurring workflow. OpenAI describes it as offering performance close to Astra at lower cost. That is the provider’s positioning, not a guaranteed saving for your account or a promise of equivalent results on every task.

Build a reusable brief: what changed, why it matters, what happens next, and which decisions need the client’s input. Evaluate whether the report remains accurate across several representative weeks—not just one easy sample.

Use Terra to work from visual material and produce clear communication

When the input is a chart, dashboard, or screen, choosing a model involves more than writing quality. It also involves reading the evidence correctly.

For a campaign review, supply the relevant screenshots and ask Terra to identify visible changes using the chart labels and dates. Tell it to flag unreadable numbers rather than guess. Then compare the output with the original dashboard.

OpenAI also positions Terra for computer use. Model capability alone, however, does not provide access to your browser, CRM, or files. Supported tools, setup, permissions, and the particular product still determine what it can do.

Give Luna a well-defined job

A focused extraction task is easier to check than an open-ended request to “handle the marketing.” Luna is a sensible candidate when the inputs and required output are consistent.

For example, ask it to turn meeting notes into a table containing an action, owner, and due date. Specify that missing owners or dates must remain marked as unspecified. Do not let a neat-looking table conceal invented commitments.

Start with clean examples, then include awkward cases: two people mentioned for one task, an ambiguous deadline, or a note that describes an idea rather than an agreed action.

Reasoning effort is a separate choice

Model selection and reasoning effort are related, but they are not the same setting. More reasoning gives a model additional opportunity to work through a problem; it can also increase waiting time and token use.

Start with the default effort available in your client. If the output misses dependencies, overlooks conflicting evidence, or needs deeper checking, try increasing effort while holding the model and inputs fixed. If the prompt omits the essential facts, more reasoning is not a substitute for providing them.

OpenAI distinguishes Max, which gives the selected model more time on a task, from Ultra, which uses subagents for separable parts of complex work. Most tasks do not need either. Luna supports effort up to Max, not Ultra; supported choices vary by model and client.

Run a small comparison before changing the whole workflow

OpenAI’s model-selection guidance recommends comparing models using the same inputs and acceptance criteria. Here is a practical way to apply that advice:

  1. Choose a real recurring task. Pick something you can judge from experience, such as a weekly report or proposal summary.
  2. Define acceptance first. List the facts that must be retained, the expected format, and the mistakes that would make the output unusable.
  3. Compare on identical material. Keep the brief, source files, and reasoning setting comparable. Change one variable at a time.
  4. Count review and repair time. Include the minutes spent checking and correcting the answer, not only the seconds before it appeared.
  5. Repeat with difficult examples. Keep the least resource-intensive configuration that consistently passes your checks.

For writing, check facts, audience fit, and useful detail. For a website or code change, inspect the result and run relevant checks. For an app workflow, reopen the affected record and verify that the change was saved without disturbing unrelated information.

The business metric is cost per usable result

A fast draft that takes substantial repair can be more expensive than a slower draft that needs only a careful review. Your effective cost includes model usage, staff review, corrections, and the consequences of errors.

Optimize for a verified result your team can use—not an impressive answer your team must rebuild. Keep your winning setup documented with a sample brief, acceptance checklist, and a review date. Re-test when the work or available models change.

If you want help turning AI experiments into reliable business workflows, explore Loudernet’s services. For a related technical foundation, see our guide to connecting coding agents to useful tools and current context.

Model names and product guidance checked November 16, 2026. Availability, controls, and limits can change; consult the models offered in your own account.

Sources

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top