OpenAI updated ChatGPT on 6 August, and while it isn't a new frontier model, two of the changes are the kind that actually alter how people use the thing day to day. One puts a reasoning dial in the hands of paying users. The other pushes a real chunk of capability down into the free tier.

A slider for how hard it thinks

The headline change for Plus and Pro users is a reasoning slider. Instead of the model silently deciding how much to think before it answers, you set it. Turn it down and you get faster, more immediate responses; turn it up and ChatGPT spends longer planning, researching, and working through harder problems — coding, multi-step questions, anything where a considered answer beats a quick one.

That's a more useful control than it might sound. The trade-off between speed and depth has always existed inside these models; making it a setting means you can match it to the task instead of taking whatever the model defaults to. A quick factual lookup doesn't need deep reasoning, and you no longer pay the latency for it; a thorny problem does, and now you can ask for it explicitly.

Alongside the slider, OpenAI says the underlying model — GPT-5.6 Sol — now gives more focused responses, makes fewer factual mistakes (it claims around 68% fewer than its previous quick-answer model, on its own internal evaluations), and cuts back on the unnecessary formatting and drift that crept into longer conversations.

The free tier gets a real lift

The more consequential change might be for people who don't pay at all. OpenAI is moving free users onto GPT-5.6 Luna as the default model, with unlimited text chats — the uncapping applies to text conversations; file uploads and image generation stay limited — and adding a "Think" button that lets free users call up higher reasoning for harder questions.

It's worth being precise about what that does and doesn't give away. Luna is OpenAI's lighter model, and the Think button makes Luna reason for longer, rather than handing free users the stronger paid model. So the free tier gets more capable, not frontier-grade. That said, it's still a real step up from the quick-answer version free users had before — the meaningful reasoning capability used to sit entirely behind the paywall, and now a slice of it doesn't. It's also a competitive move: keeping free users engaged is how you keep them from drifting to a rival's free tier.

The honest read

None of this is a leap in what the models can do — it's an iteration, and OpenAI is framing it as one, with changes rolling out over the following week. But the two changes that matter aren't about raw capability at all. They're about control and access: letting paying users decide how much thinking they want, and giving free users more of it than they had. Those are product decisions as much as model ones, and they're the sort that quietly shape how millions of people actually experience AI — which is often more important than where the frontier is this week.