Meet the models

GPT-6.2 Astra
Complex professional work
Create slides and spreadsheets that follow your templates, analyze scientific data, and design complex 3D parts.

GPT-6.2 Sol
Efficient everyday work
Write and test code, draft clear documents, and complete multistep workflows across apps.

GPT-6.1 Terra
Visual reasoning and computer use
Navigate desktop apps, write concise summaries, and fix narrowly scoped bugs.

GPT-6 Luna
Frequent, bounded tasks
Edit existing files, extract fields, and triage routine requests.
Availability, tools, reasoning settings, and usage limits differ by product and model version. Check the models available in ChatGPT or the API model catalog.
Find the right model for your workflow
Choose a task and a priority to see a suggested model and reasoning effort.
How to think about models and reasoning effort
To reduce costs or latency, start with the lightest setting that meets your acceptance criteria. Compare the API cost per completed task, including retries.
To optimize usage, start with the suggested model and reasoning effort. Check which settings your plan offers and how much of your allowance the task uses.
Model and reasoning effort
-
GPT-6 Luna · Low
Making small edits to files and extracting fields when the expected result is clear.
-
GPT-6 Luna · Extra high
Finding updates across apps, prioritizing follow-ups, and solving problems with clear constraints.
-
GPT-6.1 Terra · Medium
Navigating desktop apps, updating records, and making focused fixes you can verify.
-
GPT-6.2 Sol · Medium
Fixing bugs and adding features in an existing codebase, or drafting documents for review.
-
GPT-6.2 Sol · Extra high
Working through multistep tasks that need testing and revision along the way.
-
GPT-6.2 Astra · Low
Turning a clear brief into polished work that's ready to use.
-
GPT-6.2 Astra · Medium
Planning and carrying out complex projects, adjusting the approach as new information comes in.
-
GPT-6.2 Astra · Extra high
Solving open-ended problems that require weighing trade-offs and testing assumptions.
Farther from the center means a stronger fit for that type of work. The ratings are high-level estimates based on benchmarks, independently run task examples, and user feedback.
| Type of work | Models from strongest fit to weakest fit |
|---|---|
| 3D and design | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
| Software engineering | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
| Data analysis | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
| Creating deliverables | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
| Computer use | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
| Research decisions | GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna |
Experiment
Treat the guidance on this page as a starting point. The best way to find the right model for your workflow is to experiment with different models and reasoning settings to see what works.
Start by considering:
- How often does your workflow run? A frequent automation makes usage and cost add up faster than an occasional project.
- How quickly do you need the result? A task you’re waiting on may need a faster setting than one that runs overnight.
- How will you use the output? A draft for your review may need less polish than something you’ll share externally.
- How important is the quality of the result? Depending on your use case or industry, you might want to use a stronger model to put an emphasis on quality.
If you can, experiment using the same inputs to compare results and keep the lightest setting that meets your quality bar.
Choose acceptance criteria before comparing models:
- Writing: check required facts, qualifications, audience, and voice. Review both outputs before deciding whether the shorter one is more useful.
- Computer use: reopen the affected record or page to verify saved changes, confirm that the workflow preserved unrelated data, and compare the completion report with the actual result.
- Code and interfaces: run regression tests. For an interactive application, also inspect the rendered result on desktop and mobile.
Repeat the comparison across representative tasks. Measure completion time alongside quality, including time spent reviewing or repairing the result. Then vary reasoning effort while keeping the model and inputs fixed.