ChatGPT’s New Slider Lets You Decide How Hard It Thinks

The most annoying thing about ChatGPT has never been that it's wrong. It's that it's long. You ask a one-line question and get back a bulleted essay with three headers and a summary you didn't request. OpenAI has apparently noticed.

Alongside today's free-tier news, Plus and Pro subscribers are getting an updated GPT-5.6 Sol in ChatGPT's Chat experience, tuned to give shorter and more focused answers, and a slider that lets you set how much thought the model puts into each response. Both are live today on web, mobile, and desktop.

One dial instead of two personalities

Until now you were effectively choosing between two different experiences with two different tones: the quick one and the thinking one. That split is gone. A single model now handles both ends, and the slider decides where on the range you land, from fast everyday replies up to extended effort for planning, research, writing, and code.

This is the right shape for the problem. Nobody wants to pick a model from a dropdown before asking what time a store closes. Dragging a dial when you're about to hand it something genuinely hard is a much more natural gesture.

Less filler, more correction

The retuning is the part I'm more curious about. OpenAI says Sol now adapts its level of detail to the question, avoids unnecessary formatting, and will push back with a helpful correction rather than agreeing when agreeing wouldn't actually help you.

That last one is the interesting admission. Chatbots have a well-documented habit of telling you your idea is great. If Sol genuinely disagrees more often, that's a meaningful shift in what the product is for. If it just adds a polite caveat and then agrees anyway, nothing has changed except the word count.

On accuracy, OpenAI's internal evaluation says responses with at least one factual error are 68 percent less common with the updated Sol than with GPT-5.5 Instant. Internal numbers, self-reported, on a benchmark the company designed. Useful signal, not proof.

Read the fine print on scope

Here's the part that will confuse people. This version of Sol is tuned specifically for everyday conversation, so it only runs in the Chat experience. The Sol that powers ChatGPT Work and Codex is not changing in this release.

So if you spend your day in Codex and were hoping for tighter, less padded output there, this isn't your update. Same model name, different build, different behavior. Model naming across the industry remains a mess, and this doesn't help.

Who should care

Anyone paying $20 a month who has quietly developed a habit of skimming past the first three paragraphs of every answer. If the retuning holds up, the practical win isn't smarter responses. It's shorter ones that respect the size of the question you actually asked.

That's a smaller promise than a new model generation, and probably a more useful one.