ChatGPT Models
ChatGPT's 3 New Models Explained Simply
Sol, Terra and Luna each have a completely different job. Here is when to use each one, where to find them and how to stop burning your usage on the wrong model.
Back to resourcesI have been cheating on Claude with ChatGPT's new models all week.
OpenAI has released three versions of GPT-5.6: Sol, Terra and Luna. The names sound similar, but they are not three interchangeable buttons. Each one is built around a different trade-off between intelligence, speed and cost.
The easiest way to remember them is this: Sol for your hardest work. Terra for everyday work. Luna for fast work.
The Simple Version
Think Power, Balance, Speed
- Sol = power. Use it when getting the strongest answer matters more than getting the fastest answer.
- Terra = balance. Use it when you want strong everyday work without paying or waiting for the flagship model.
- Luna = speed. Use it for quick, repeatable or high-volume jobs where efficiency matters most.
Sol: When You Need The Best Answer
Sol is the flagship and most capable model in the GPT-5.6 family. It is designed for complex professional work: deep research, strategy, difficult coding, knowledge work and multi-step problems.
I have been reaching for Sol any time I really need to get the job done and I am happy to wait for a better answer.
Use Sol for:
- deep research that needs synthesis, not just a summary
- strategy, scenario planning and difficult decisions
- complex coding, debugging and technical architecture
- long documents or projects with several moving parts
- a final review where accuracy and judgment matter
Do not waste Sol on: rewriting one sentence, extracting a date, formatting a list or drafting a routine reply. It can do those jobs, but that does not make it the right tool.
Terra: The Everyday Model
Terra is the middle of the family. OpenAI describes it as the balance between intelligence and cost, which makes it the sensible starting point for most real work.
This is the model to use when a task still needs judgment, but does not need the full weight of Sol.
Use Terra for:
- emails, reports and first drafts
- summarising documents and meetings
- content planning and everyday analysis
- routine coding and website changes
- turning research into a useful deliverable
If you are not sure which model a task needs, start with Terra. Move up to Sol when the task is genuinely difficult. Move down to Luna when the process is clear and speed matters more.
Luna: When Speed Is The Job
Luna is the fastest and lowest-cost model in the GPT-5.6 family. It is designed for cost-sensitive, high-volume work.
That does not mean it is only useful for throwaway tasks. It means it is a strong choice when the job is clear, repeatable and easy to check.
Use Luna for:
- quick summaries and information extraction
- classification, tagging and sorting
- formatting or restructuring existing text
- simple code edits and repetitive checks
- large batches of similar tasks
The smaller the decision and the clearer your instructions, the more sense Luna makes.
Why The Efficiency Matters
If you have ever hit a usage limit in the middle of a project, you already know why model choice matters.
In the API, Sol is priced at $5 per million input tokens and $30 per million output tokens. Terra is half that price at $2.50 and $15. Luna is one-fifth of Sol's price at $1 and $6. All three support the same 1.05 million-token context window, but they are designed to spend that capacity differently.
Independent benchmark results also put Sol very close to Claude's strongest model on a broad intelligence index, while some task-level comparisons show a much lower estimated cost. Treat that as evidence of efficiency, not a promise that every Sol task will cost exactly one-third as much. Your prompt, reasoning level and output length can change the real cost dramatically.
Where You Can Actually Find Each Model
This is the part most explainers miss: you cannot select all three models in a normal ChatGPT conversation.
- Standard ChatGPT chats: eligible paid plans can use Sol through the Medium, High, Extra high and Pro reasoning options. Terra and Luna are not selectable here.
- ChatGPT Work: eligible Plus, Pro, Business and Enterprise plans can choose Sol, Terra or Luna.
- Codex: eligible paid plans can use all three. Free and Go plans can use Terra.
- OpenAI API: developers can use Sol, Terra and Luna directly.
Reasoning Level Changes The Equation
The model is only half the choice. You can also change how much reasoning effort it uses.
A higher reasoning setting gives the model more room to work through a difficult problem, but it also takes longer and uses more of your allowance. Start with the lowest setting that fits the task. Increase it when the first answer is not good enough or the consequences of a weak answer are high.
- Low or none: clear, routine and easy-to-check work
- Medium: most professional tasks
- High or above: difficult analysis, coding or multi-step reasoning
The Cheat Sheet
Choose In Ten Seconds
- I need the strongest answer. Choose Sol.
- I need a strong everyday result. Choose Terra.
- I need it fast, cheap or at scale. Choose Luna.
- I am unsure. Start with Terra, then move up or down.
- The result really matters. Use Sol and give it more reasoning.
- The task is repetitive and easy to verify. Use Luna.
You do not need one favourite model. You need a simple escalation rule: Luna for speed, Terra by default, Sol when the stakes or complexity justify it.
Use The Models In Real Work
See how Chat, Work, Codex and Sites fit together, then explore six practical things you can create with ChatGPT Work.