# What is GPT-5.1, and what changes in ChatGPT's Instant and Thinking models?

Date: 2025-11-13 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-is-gpt-5-1/

**Answer:** GPT-5.1 is the update OpenAI released on November 12, 2025, to the two main models behind ChatGPT, GPT-5.1 Instant and GPT-5.1 Thinking. Instant is warmer, follows instructions more closely and can now decide to think before hard questions, Thinking adjusts how long it reasons to each question, and ChatGPT gains new tone presets such as Professional, Candid and Quirky.

Yesterday, Wednesday, November 12, 2025, OpenAI upgraded the models inside ChatGPT to **GPT-5.1**. It is not a new generation: OpenAI says the ".1" signals real improvements while staying in the GPT-5 family, and that future step-by-step upgrades will be named the same way. Most people will notice it in two places, how ChatGPT sounds and how long it takes to answer.

The update comes in two versions. **GPT-5.1 Instant** is the fast model that handles most everyday chats. **GPT-5.1 Thinking** is the reasoning model that works through hard problems before replying. Alongside them, ChatGPT gets an expanded set of tone and style controls.

## What OpenAI released

| | GPT-5.1 Instant | GPT-5.1 Thinking |
|---|---|---|
| Role | Default everyday model | Advanced reasoning model |
| Main change | Warmer, more conversational, better at following instructions | Spends time in proportion to the difficulty of the question |
| New ability | Can choose to think briefly before a harder question | Clearer answers with less jargon |
| API name | gpt-5.1-chat-latest | GPT-5.1 |

The rollout begins with paid plans (Pro, Plus, Go and Business), then reaches free users and people who are not logged in. OpenAI is spreading it over several days to keep the service stable, so some people won't see it right away. Enterprise and Education plans get a seven-day early-access switch before GPT-5.1 becomes the only default. Both models are due in the API, the paid service developers use to build apps, later this week.

GPT-5 is not removed overnight. Paid subscribers can still pick it from the legacy models menu for three months, a period OpenAI says gives people time to compare the two at their own pace.

## How adaptive reasoning works

A **reasoning model** writes out hidden working steps before its final answer, the way a student does scratch work before writing a result. That thinking improves accuracy on math, code and logic, but it costs time. The question is always how much to think.

GPT-5.1 makes that choice more flexible in two ways:

- **Instant** can now use what OpenAI calls **adaptive reasoning**: for most messages it replies at once, but for a harder question it can decide to pause and think first. OpenAI reports better results on math and coding tests such as AIME 2025, a competition math exam, and Codeforces, a programming contest site.
- **Thinking** varies its effort more than before. On a typical mix of ChatGPT requests, OpenAI says it is about twice as fast as GPT-5 Thinking on the easiest tasks and spends about twice as long on the hardest.

Most people won't have to choose a model at all. **GPT-5.1 Auto** keeps routing each message to whichever model suits it, much like a receptionist sending simple questions to the front desk and complex ones to a specialist.

OpenAI also says Instant follows instructions more reliably. In its own example, a user asks for answers of exactly six words, and GPT-5.1 Instant keeps to that rule over several turns where GPT-5 drifted.

## Personality presets and finer controls

The second half of the release is about tone. ChatGPT already had a few preset styles. OpenAI is renaming some and adding others:

| Preset | Status |
|---|---|
| Default | Updated |
| Friendly | Formerly "Listener" |
| Efficient | Formerly "Robot" |
| Professional, Candid, Quirky | New |
| Cynical, Nerdy | Kept unchanged |

Ars Technica's report explains that these presets work by changing the instructions placed in front of each conversation, the **system prompt**; the model's abilities are the same whatever style you pick. OpenAI is also testing, with a limited group of users at first, controls for how concise, warm or easy to scan replies are, and how often emojis appear. If you keep asking for a particular style in a chat, ChatGPT may offer to save it as a setting.

One practical change: edits to personalization now apply to every chat right away, including conversations already in progress. Before, they only affected chats you started afterward.

> **Try it:** Ask the same question twice, once with the Efficient preset and once with Friendly, for example "Explain what a mortgage rate is." Compare length, tone and whether the facts stay the same.

## Why it matters

This release says a lot about where OpenAI's attention is. Earlier this year it rolled back a model update for being too flattering, and later other users objected when OpenAI changed GPT-5's default style. Ars Technica describes GPT-5.1 as an attempt to balance those opposite complaints by letting each person set the tone.

Fidji Simo, OpenAI's CEO of Applications, pointed out that ChatGPT now has over 800 million users, so a single style cannot suit everyone. She also named the risk on the other side, as quoted by Ars Technica: "Personalization taken to an extreme wouldn't be helpful if it only reinforces your worldview or tells you what you want to hear." OpenAI published a system card addendum on its safety approach for the new models.

For everyday users, the changes are practical: quicker answers to simple questions, more thorough ones on hard problems, and an easier way to get the voice you prefer.

## How do you set ChatGPT's tone with the new personality presets?

Open ChatGPT's settings and go to **Personalization**. There you will find the style and tone menu with the presets listed above. Pick one, and it applies at once to all your chats, including any that are open. If you have custom instructions, keep them: OpenAI says GPT-5.1 follows them more closely, so a line such as "Use plain English and define technical terms" now carries more weight.

Choose by task rather than by mood. Efficient suits quick lookups and checklists; Professional fits work email and reports; Friendly or Candid suit brainstorming. Because a preset only changes instructions, not the model's knowledge, switching style never makes answers more or less accurate. Check important facts either way.

If you are among the users testing the finer controls, try one change at a time, for instance fewer emojis, so you can tell what each setting does. You can remove any preference later, and ChatGPT may suggest an adjustment when you repeatedly ask for shorter or longer replies.

## Sources

- GPT-5.1: A smarter, more conversational ChatGPT: https://openai.com/index/gpt-5-1/
- OpenAI walks a tricky tightrope with GPT-5.1's eight new personalities: https://arstechnica.com/ai/2025/11/openai-walks-a-tricky-tightrope-with-gpt-5-1s-eight-new-personalities/