← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

ChatGPT’s Models,
Explained Simply

Sol, Terra and Luna started with GPT-5.6. GPT-6 kept the same names and added Astra on top. Here’s what each one is for, and when to reach for which.

When OpenAI shipped GPT-5.6 in July, the smart part was the packaging. It wasn’t one model that tries to do everything. It was three, each built for a different kind of job.

Here’s the idea that makes it click. The number is the generation, so it tells you how new the tech is. The name is the tier, so it tells you how much power you’re paying for.

That’s exactly what happened next. GPT-6 arrived in the fall with a Sol, a Terra and a Luna, plus a new top tier called Astra. Same jobs, newer brains. Some tiers are already on 6.1, so you’ll see names like GPT-6.1 Sol.

What They Cost

These are OpenAI’s API prices per 1 million tokens today. Input is what you send in, output is what it writes back. Every tier reads about 1 million tokens of context and writes up to 128,000 in one go. (OpenAI’s model docs)

  1. GPT-6 Astra: $10 in / $50 out. The most capable.
  2. GPT-6.1 Sol: $2 in / $10 out. Near-Astra for complex work.
  3. GPT-6.1 Terra: $1 in / $5 out. The fast, efficient one.
  4. GPT-6 Luna: $0.10 in / $0.50 out. The cheapest by far.

The GPT-5.6 versions are still around and now cost $4/$20 (Sol), $2/$12 (Terra) and $0.20/$1.20 (Luna). The newer tiers are cheaper, so there’s little reason to start a new project on 5.6.

What Each One Is For

01

Astra, the top tier

The top of the line for the hardest work, like complex reasoning, coding, research and big documents. In ChatGPT it shows up as GPT-6 Pro on the higher plans. You don’t reach for it to clean up an email.

02

Sol, the powerhouse

OpenAI describes 6.1 Sol as near-Astra performance at a lower cost, for complex coding, computer use and professional work. It’s the one I reach for most. For scale, it costs half what Claude Opus 5.5 does in the API (Anthropic’s pricing), and cheaper thinking means more thinking before you hit a limit.

03

Terra, the workhorse

Fast enough to run at volume, sharp enough to trust. OpenAI built 6.1 Terra for clear writing, visual reasoning and precise computer use. Right now it lives in ChatGPT Work and Codex, not the regular chat.

04

Luna, the fast one

The fastest and cheapest. Give it summaries, first drafts, sorting and routine automation. It won’t out-think Sol and isn’t trying to. It’s also the GPT-6 model Free and Go users get in chat.

Which One, When
  1. A wrong answer would cost real money: Astra.
  2. Deep research, complex coding, strategy: Sol. Best value for hard work.
  3. Everyday writing, document work, agent tasks at volume: Terra.
  4. Quick, simple, high-volume: Luna. Run it all day.

In the ChatGPT app you mostly pick GPT-6 and your plan decides the tier. Paid plans get Sol, Free and Go get Luna, and the 5.6 models stay in the picker under their own family. Not sure which tier a task needs? Paste this first.

Pick My Model Tier
You are my AI model router. Before I start a task, tell me which OpenAI model tier to use so I don't burn my best model's limits on easy work or get a weak answer on hard work. The tiers, strongest to fastest: - Astra: the most capable. For work where a wrong answer is expensive. - Sol: close to Astra on complex work for much less. My default for hard tasks. - Terra: fast and efficient for clear writing, visual reasoning and computer use. - Luna: the cheapest and fastest. For simple, high-volume jobs. My task: [describe it, e.g. "rewrite 40 product descriptions in our brand voice"] How often I'll do it: [once / daily / thousands of times through the API] What a mistake would cost: [low / medium / high] Where I'm working: [ChatGPT chat / ChatGPT Work / Codex / the API] Do this: 1. Pick one tier and give me 2 reasons in plain words. 2. Name the cheaper tier I could try first, and the sign that tells me to step up. 3. If I'm in ChatGPT and that tier isn't in my model picker on my plan, say so and tell me the closest option, or tell me to check the picker if you're not sure. 4. Rewrite my task as a clear first prompt for that model: the goal, the inputs, the steps and the exact output format. Rules: don't recommend the biggest model by default. Keep your answer under 200 words plus the rewritten prompt.

The Real Skill

Think of AI models like streaming services. If you only have one, you miss a lot. Knowing which one to open for the job in front of you is the actual skill, and OpenAI’s naming makes that easy.