ChatGPT gets an effort slider: you decide how hard it tries

What happened: On August 6, 2026, OpenAI rolled out a new model, GPT-5.6 Sol, to ChatGPT Plus and Pro users. The headline feature is an effort slider right in the chat window on web, mobile and desktop — it sets how much “thinking” the model puts into a given answer. In OpenAI's internal test on financial, legal and medical prompts, GPT-5.6 Sol produced about 68% fewer answers with at least one factual error than the older GPT-5.5 Instant. The test methodology wasn't published publicly; it's OpenAI's own measurement. The model replaces GPT-5.5 Instant as the default on paid plans; a similar slider previously existed only in ChatGPT Work and is now reaching regular paid accounts too. At the same time, GPT-5.6 Luna became the default model for free users, who are also gradually getting unlimited text chats.

What it means for you: Until now, the speed-versus-quality tradeoff in ChatGPT was handled indirectly — you either waited for a “thinking” model or got a fast but shallower answer. Now you set it per question:

• Low setting for routine things — a quick email summary, a draft reply, a simple factual question.
• High setting for decisions, planning, writing or code where it pays off for the model to think the answer through — a next-quarter budget draft, or reviewing a contract before you sign it.

In practice: before asking it to analyze numbers from a report or draft a strategy, push the slider up — you'll get an answer that weighs more angles, at the cost of a longer wait. For repetitive routine tasks (rephrase a paragraph, translate a sentence, find a date in your calendar), paying extra time isn't worth it — leave the slider low and the answer arrives almost instantly.

How to start:

1. In ChatGPT (web, mobile or desktop), open a new chat on a Plus or Pro plan.
2. Next to the message box or in the model settings, find the effort slider.
3. Leave it low for routine questions; raise it for decisions that matter.
4. Watch how response time changes — higher effort means a longer wait, but also a more thorough answer.

What to watch out for: The 68% improvement figure comes from OpenAI's own internal test, not independent verification — treat it as directional, not a guarantee. Even at high effort the model can still get things wrong, so for facts that matter (numbers in a contract, legal or medical information) always verify the answer against a primary source before relying on it at work. And before you paste a whole contract or company report into ChatGPT for analysis, think about what's in it — personal data, client information or internal figures don't belong there without your company's approval; anonymize sensitive parts first.