ChatGPT just renamed its brain: meet Sol

Hi,

Two things this week: what OpenAI's new Sol model actually means for you, and a prompting habit that costs nothing and improves almost every result.

What is Sol, and should you care?

OpenAI released GPT-5.6 in July, and with it a new naming system. Instead of one model with a confusing version number, there are now three tiers: Sol, Terra and Luna.

Think of it like cabin classes on a flight. Sol is the flagship, built for complex work such as research, coding and long documents. Terra is the balanced middle option for everyday tasks. Luna is the fastest and cheapest, for quick and simple jobs.

A few things worth knowing:

• Sol is only available on paid ChatGPT plans. Free users get Terra, and only inside the new ChatGPT Work and Codex tools.
• The names are here to stay. When GPT-5.7 arrives, it will still be Sol, Terra and Luna. That makes it easier to know what you are choosing.
• You can pair each model with a reasoning setting. More reasoning means better answers on hard problems, but slower responses and faster use of your limits.

My simple rule: start with the best model you have on the lowest reasoning setting. Only turn the reasoning up if the answer disappoints you. Most everyday tasks do not need the expensive thinking.

If you want the detail, OpenAI's announcement is here.

One prompting tip: give the model an out

Here is a habit I use in almost every serious prompt. Add one line at the end:

"If any part of this is unclear or you are missing information, ask me before answering."

Why it works. AI models are trained to answer, even when your request is vague. That is how you get confident, polished responses built on wrong assumptions. Giving the model explicit permission to ask questions first turns a guessing exercise into a conversation.

I use this constantly at work. When I ask for a summary of a document, a draft email or an analysis, that single line regularly surfaces a question I had not thought to answer. The final output is better because the model understood the job before starting it.

Try it this week on one real task. You will notice the difference in the first reply.

That is it for this issue. Reply if you try either of these, I read every response.