Just now, GPT-5.1 was officially released. There's something "off" about OpenAI this time.
Just now, OpenAI officially released GPT-5.1, but this time it's a bit different.
After going through the entire official blog, I found a really interesting detail: There isn't a single score comparison chart. There's no benchmark data, no "XX% performance improvement", and even the common catchphrase "faster and stronger" is hardly seen.
This doesn't seem like OpenAI's style. It wasn't until I saw this sentence:
“We've clearly heard from users that a great AI not only needs to be smart but also make chatting with it a pleasant experience.”
The new GPT-5.1 offers eight preset styles to make our chats more enjoyable.
The new version is indeed smarter - with more rigorous reasoning and better - written code. But what's most notable is that it finally seems more human, and for the first time, allows us to finely "tune" its chatting style.
Chatting with AI is no longer just a question - and - answer tool - like experience. It has become full of jokes, understands the atmosphere, can pick up on the conversation, and can even chat with you about all sorts of things for a long time.
It seems that after the poor reception of GPT-5 last time, OpenAI finally listened to the advice and for the first time broke the ice, admitting that just topping the charts is useless. Users want an AI that can communicate well, providing both practical value and emotional value.
Here are the specific usage times and methods: After updating to GPT-5.1, our ChatGPT will automatically switch to the latest model without the need for special selection.
Paid users (Pro, Plus, Go, Business): The update will be gradually pushed starting today (November 12th).
Free and unregistered users: They will receive the update after the paid users have all been updated.
Enterprise and education versions: There is a 7 - day early - bird switch (off by default), after which they will be uniformly upgraded to GPT-5.1.
API developers: GPT-5.1 Instant and GPT-5.1 Thinking will be available on the API later this week.
A More Powerful AI Core
The core of this update is the full - scale upgrade of the two major models, GPT-5.1 Instant and GPT-5.1 Thinking.
GPT-5.1 Instant: The Most Commonly Used Model Gets "Warmer"
GPT-5.1 Instant is the most frequently called model in ChatGPT. This time, it has become "warmer" and more talkative. According to OpenAI's early tests, it even shows a bit of playfulness from time to time while keeping its answers clear and useful.
The more crucial upgrades come from the underlying level:
1. More Obedient: It can now more reliably follow our instructions and accurately answer the questions we really want to ask.
2. Adaptive Reasoning: This is the first time this function has been introduced to the Instant model. It means that when facing difficult problems, it will intelligently decide to think first, thus giving more thorough and accurate answers; while for simple questions, it still responds extremely quickly.
OpenAI mentioned that this evolution has also shown significant improvements in professional evaluation test sets such as mathematics (AIME 2025) and programming (Codeforces).
GPT-5.1 Thinking: A Stronger Brain and More Understandable
As a more advanced reasoning model, GPT-5.1 Thinking has also undergone key optimizations, becoming more efficient and user - friendly.
GPT-5.1 Thinking spends less time on simple tasks and more time on difficult tasks.
1. Efficiency Improvement: It can now more accurately allocate thinking time, spending more time on complex problems (for more thorough answers) and responding faster to simple problems (with shorter waiting times).
2. More Understandable (A Boon for Users!): Its answers are now clearer, using fewer industry jargons and undefined words. This allows us to easily understand when using it to handle complex work or explain technical concepts.
3. Equally Warm: The default tone of the Thinking model has also become warmer and more empathetic.
In the words of Fidji Simo, the CEO of OpenAI Applications, the core of this upgrade is to better combine IQ (intelligence quotient) and EQ (emotional quotient).
While maintaining a high IQ, that is, continuing to use the same technology stack as the reasoning model, the model has also significantly improved its EQ. ChatGPT now has more natural conversations and empathy. This can meet users' corresponding personalized needs in different scenarios, such as the need for empathy when talking about health and directness when writing copy.
Moreover, for most users, we don't need to struggle between Instant and Thinking. Because there is also GPT-5.1 - Auto, which will automatically assign us to the most suitable model. This was also a major highlight when GPT-5 was released, namely intelligent routing.
In short, the most intuitive feeling is that the answers are smarter and the tone is more natural.
Create Your Own ChatGPT
If the model upgrade is hard power, then the leap in personalized experience is soft power, and this is another major highlight of this update.
OpenAI's goal is to allow us to effortlessly adjust ChatGPT's tone and style to the most comfortable state.
On top of the existing default, friendly, and efficient styles, three new official styles have been added.
Professional: Suitable for formal occasions such as work and writing.
Candid: More direct and to the point.
Quirky: As the name suggests, it will become more interesting and lively.
The "tech geek" and "complaint expert" options from the previous beta version are still retained in the personalized settings.
In addition to directly selecting these styles, with a richer range of basic styles and tones, OpenAI is experimenting with a new feature, allowing users to fine - tune ChatGPT's characteristics directly from the settings.
We can precisely control the conciseness of the answers, the enthusiasm level, whether the answers are scannable, and even the frequency of using emojis.
If you don't want to bother with manual settings, when we try to guide a certain tone in the conversation, ChatGPT may even proactively understand it and then ask us if we want to save this preference in the permanent settings, saving us the trouble of manual adjustment.
This is also a feature that Altman likes.
In the blog shared by Fidji Simo, she mentioned that the previous custom instructions were not always satisfactory. For example, we could tell ChatGPT not to use a certain word in the custom settings, but it still did.
Another major improvement in the stylization of GPT-5.1 is that custom instructions can now more reliably hold up in multi - round conversations. ChatGPT can more stably complete various tasks according to the personality we define.
Some netizens summarized it in one sentence: The update of GPT-5.1 is to more creatively ignore our prompt words.
Of course, the stylization and anthropomorphization of AI also come at a cost. A warmer and more emotionally intelligent AI must also be safer. This is also something that OpenAI must address after being involved in the suicide case of a 16 - year - old boy recently.
In the System Card introduction of the GPT-5.1 model, it describes OpenAI's in - depth considerations in this regard. For the first time, OpenAI has added two brand - new and more human - oriented dimensions to the model's security assessment.
1. Mental Health: Assess how the AI responds to signs such as isolation, delusion, or mania that users may show.
2. Emotional Reliance: Assess whether the AI's responses will encourage users to develop an unhealthy dependence or emotional attachment to ChatGPT.
In traditional security assessments, GPT-5.1 Instant performs excellently and is stronger than its predecessor, gpt - 5 - instant - oct3, in resisting jailbreaks.
However, OpenAI also candidly pointed out that GPT-5.1 Thinking has slightly regressed in benchmark tests for handling harassment, hate speech, etc., compared to its predecessor; the Instant model also shows a slight regression in some evaluations of emotional reliance.
OpenAI, of course, said that it is committed to improving these aspects and mentioned that they chose to transparently disclose this regression phenomenon. In the current era of rapid AI iteration, this is more worthy of our attention than simply claiming zero errors.
Some netizens also said that they are glad to see that OpenAI is making efforts to make our conversations with the model more enjoyable.
If you don't see the update when you open ChatGPT today, don't worry. It will be your turn in the next few days as OpenAI is gradually pushing the update to all users.
Moreover, to avoid the embarrassment that happened when GPT-5 was released last time, when netizens were calling for the return of GPT - 4o.
OpenAI is offering a "second chance" this time. Paid users can still select the old GPT-5 model from the drop - down menu in the settings within 3 months, which is the phasing - out period of GPT-5, so that they can smoothly transition to GPT-5.1.
You can still use models like 4o for now.
GPT-5.1 is a major update that improves both capabilities and user experience.
OpenAI is clearly telling us that the future of AI is not just about stronger parameters and higher scores, but also about a more user - understanding experience and more considerate interaction.
But what should a perfect assistant be like?
Fidji Simo, the CEO of OpenAI Applications, made an interesting analogy in her article. She said, "If I could completely control my husband's traits, I might make him always agree with me, but obviously, that's not a good idea."
The best AI should be like the best people in our lives. They listen, adapt, but also challenge us when necessary to help us grow.
From an all - knowing tool to a personalized partner that understands your chatting style and can even help you grow, this may be the future that GPT-5.1 really wants to open up.