Everyone's posting benchmark charts. Skip them. Here are the only three things worth doing with OpenAI's new model, each with a copy-paste prompt you can run in the next five minutes.
What's GPT-5.6? It's OpenAI's newest family of models. The top one is called Sol, with cheaper versions named Terra and Luna. Two things changed that a normal person should care about: it got cheaper, and it can now operate your screen for you instead of just answering questions.
Terra is roughly as good as the previous top model, for about half the price. For most of what you do (drafting, summarising, research) you don't need the most expensive model. Set the cheaper tier as your default and save the top one for the genuinely hard stuff.
Run my next normal task on the cheaper tier and see if I can even tell the difference. Draft me a reply to this email: [paste email].
This is the big one. It can now click through your screen and do things, not just answer. Start small so you can watch it work, and keep it checking in with you before anything final.
I want you to do this on my screen, step by step, and check with me before anything final: find the cheapest flight from Sydney to Melbourne next Friday morning, get it to the checkout page, then stop and let me confirm.
The shift this year is from asking to delegating. Instead of one question at a time, give it a multi-step job and let it run the whole thing.
Here's a job, not a question. Do all of it:
1. [step one]
2. [step two]
3. [step three]
Then show me the finished result and flag anything you weren't sure about.