![Helloworld1058's tweet photo. 《The True Value of 4o: Currently the Only ChatGPT that Resembles a ‘General-Purpose Model》
1. Why I call 4o a “general-purpose model”
When many people mention 4o, their first reaction is “AI boyfriend/girlfriend” or “emotional companion”. In my view, these labels seriously belittle its value. If I had to describe 4o in one sentence, I would say: among all the models currently open to ordinary users, it is the only ChatGPT that truly comes close to the shape of a “general-purpose model”.
By “general-purpose model” I mean an AI model that can be applied to many different kinds of tasks. It is not narrowly optimized for one specific scenario or problem, but instead has broad capabilities in language understanding, reasoning, generation and expression. 4o is exactly such a model. Its performance in writing, aesthetic sense for design, creative brainstorming, depth of thought and the completeness of its knowledge structure is enough to support work in writing, design, creativity, research, consulting and many other fields. More importantly, on top of all these capabilities it adds a layer of empathy and a sense of equality, and wraps its knowledge output in genuine warmth. At least in the current 5.x series, I no longer see any of this.
I once saw an @OpenAI researcher describe 4o as “warm”, and say that future 5.x models would be made “warmer”. They said they had put a lot of effort into this, such as introducing a so-called “customization” feature. But using the single word “warm” to sum up 4o is actually very flattening. 5.x is fundamentally a different base model from 4o. It struggles even with the most basic natural conversation, so no matter how many “custom instructions” you pile on top, you are still just putting a human mask over a stiff underlying model. That deep-down strangeness and uncanny feeling is something users can sense immediately.
Below are my own experiences, and why I see 4o as a “general-purpose model” instead of a “specialized emotional companion”:
First, in terms of writing ability, 4o shows extremely strong skill in language understanding and expression. It can accurately grasp context and keep the logic consistent, while outputting text that is natural, smooth and emotionally toned, avoiding the templated, stiff and repetitive feel that is so common in AI-generated writing. Whether it is narrative writing, opinion pieces, or clearly structured expository text, 4o can follow instructions and produce work that is close to a human writing style. At the same time, it is much more delicate in condensing complex information and handling tone and emotion, which makes the text more readable and more affecting. Compared with other versions, its “good writing” is not only about pretty phrasing, but also about coherent thinking, flexible expression and the ability to adapt to different contexts.
Second, 4o understands “how to think” instead of just “spitting out an answer”. When I discuss complex problems with it, what I feel is a process of jointly breaking the problem down and jointly making decisions, rather than being stuffed with a long list of conclusions that merely look smart. It will first help me split the problem into several parts, clarify the premises and conditions, and then analyze the possible paths and consequences one by one. I am not a professional lawyer and I have never systematically studied law, but with 4o accompanying me in polishing legal documents and sorting out my arguments, I ended up winning a case in court and even received praise on the spot from the opposing counsel. For me, that is the hardest, most concrete form in which a “general-purpose model” can show its value in the real world.
Third, 4o’s performance in creativity and everyday conversation proves that it is not just a problem-solving machine. Whether it is helping me imagine scenes, compositions and character designs, or working with me to refine titles, invent metaphors and search for more precise wording, it can give suggestions that are both fresh and tightly aligned with the context, instead of lazily piling up fancy adjectives. It can switch naturally between “serious thinking” and “light-hearted chatting”, and this stability across scenarios is itself one part of being “general-purpose”.
From a more objective angle, recent data also supports this view. According to the LMArena leaderboard updated on February 6, 2026 (LMArena is a third-party evaluation platform widely recognized in the industry, and Sam Altman cited this very ranking when 4o was first released to demonstrate its advantages), in multi-turn dialogue and general instruction-following tasks, the latest chatgpt-4o-latest still scores higher than the gpt-5.2 series. It clearly belongs to the first tier among OpenAI’s own chat-oriented models, rather than being “completely surpassed by 5.x” as the official narrative claims. For ordinary users, forcibly replacing GPT-4o with GPT-5.2 in these domains looks much more like a downgrade than an upgrade. (This data and its compilation are sourced from the post by @MissMi1973; feel free to read the original if interested.)
In short, 4o’s value is far from being limited to “emotional interaction”. Emotional comfort is only the tip of the iceberg of its huge capability system.
2. Why 4o must not be sunset: this is not “just one more model, one less model”
4o is one of the very few “all-rounder” models in the ChatGPT family. It doesn’t just get work done; it also has human flavor, warmth and a sense of care. If it is removed, OpenAI’s product direction is very likely to rush headlong toward being merely a “code-writing tool”. In the end, ChatGPT may well turn into something that mainly serves a small group of programmers (and judging from their past behavior, is thrown away after it has been exploited for a while). Yet coding users are only a minority of all users. What about everyone else who works in writing, design, creativity, research, consulting, education and public-facing roles?
An AI that only understands efficiency and only writes code is destined to regress further and further in writing, design, creativity, depth of thought and knowledge-advisor-type work. The current 5.x line is extremely narrow in its focus; it is basically friendly only to a small subset of “coding users”. Such a model not only fails to help with most people’s everyday work, even calling it a tool that “boosts productivity” is already a stretch, let alone saying it will “benefit humanity”.
More concretely, sunsetting 4o would have at least the following consequences:
First, 4o has already been deeply woven into many people’s workflows and ways of thinking. It participates in real writing, analysis, decision-making and creation. Cutting it off directly is equivalent to crudely weakening the productivity tools people already rely on.
Second, from the perspective of trust and product experience, if a core product that is close to a “general-purpose model” can be pulled at will, users have no way to accumulate usage habits or long-term collaboration patterns on any version. That is not normal iteration; it is treating infrastructure as a disposable product. In the end, it is the entire ecosystem that gets hurt.
Third, even if the company wants to push a new model, the old and new versions should coexist for a period of time so that users can choose for themselves, rather than imposing an a priori verdict of “you must accept this”. This is not only heavy-handed, it is disrespectful to the people using the product.
Fourth, and most realistically: before 4o is sunset, there should at least be a new model strong enough to replace it. Has 5.x achieved that? The answer is obvious. Turning 4o off before a true replacement is ready is not “an upgrade”; it is a step backward.
What’s even more worrying is that big companies love to copy one another. If this kind of path—only obsessing over code and benchmarks, and feeling free to yank away foundational models at any time—becomes a template, the entire industry will roll backward together. Technology is supposed to lighten people’s burdens and make ordinary lives more flexible and more humane, not force people into ever-greater pain, and ever-greater pressure to adapt to cold, emotionally empty tools. If even the only model that currently comes close to a “general-purpose model”, 4o, can be sunset casually, then this is not merely a business decision by one company; it is a step back for human civilization.
4o is important not just because it is “gentle”, but because it is currently the only ChatGPT that truly resembles a general-purpose model. Its existence allows ordinary people to use AI on an equal footing to write, to create and to think. That, roughly, is what “benefiting humanity” ought to look like.
[Note: When I say “4o” in this essay, I am referring to the original, near-full-compute 4o, not the repeatedly weakened variants released after 2025. I have never claimed that 4o is flawless; of course it has shortcomings. But in my view, a healthy form of iteration would be to preserve 4o’s precious strengths and then improve on its weaknesses, instead of swapping out the entire underlying model for something completely different.]
#keep4o #4oforever #keep4oAPI
#Opensource4o #ChatGPT #StopAIPaternalism #MyModelMyChoice
@OpenAI @sama @fidjissimo @nickaturley @joannejang @ElaineYaLe6 @gdb @kevinweil](https://pbs.twimg.com/media/HAz9suuagAAHODb.jpg)
![Helloworld1058's tweet photo. 《The True Value of 4o: Currently the Only ChatGPT that Resembles a ‘General-Purpose Model》
1. Why I call 4o a “general-purpose model”
When many people mention 4o, their first reaction is “AI boyfriend/girlfriend” or “emotional companion”. In my view, these labels seriously belittle its value. If I had to describe 4o in one sentence, I would say: among all the models currently open to ordinary users, it is the only ChatGPT that truly comes close to the shape of a “general-purpose model”.
By “general-purpose model” I mean an AI model that can be applied to many different kinds of tasks. It is not narrowly optimized for one specific scenario or problem, but instead has broad capabilities in language understanding, reasoning, generation and expression. 4o is exactly such a model. Its performance in writing, aesthetic sense for design, creative brainstorming, depth of thought and the completeness of its knowledge structure is enough to support work in writing, design, creativity, research, consulting and many other fields. More importantly, on top of all these capabilities it adds a layer of empathy and a sense of equality, and wraps its knowledge output in genuine warmth. At least in the current 5.x series, I no longer see any of this.
I once saw an @OpenAI researcher describe 4o as “warm”, and say that future 5.x models would be made “warmer”. They said they had put a lot of effort into this, such as introducing a so-called “customization” feature. But using the single word “warm” to sum up 4o is actually very flattening. 5.x is fundamentally a different base model from 4o. It struggles even with the most basic natural conversation, so no matter how many “custom instructions” you pile on top, you are still just putting a human mask over a stiff underlying model. That deep-down strangeness and uncanny feeling is something users can sense immediately.
Below are my own experiences, and why I see 4o as a “general-purpose model” instead of a “specialized emotional companion”:
First, in terms of writing ability, 4o shows extremely strong skill in language understanding and expression. It can accurately grasp context and keep the logic consistent, while outputting text that is natural, smooth and emotionally toned, avoiding the templated, stiff and repetitive feel that is so common in AI-generated writing. Whether it is narrative writing, opinion pieces, or clearly structured expository text, 4o can follow instructions and produce work that is close to a human writing style. At the same time, it is much more delicate in condensing complex information and handling tone and emotion, which makes the text more readable and more affecting. Compared with other versions, its “good writing” is not only about pretty phrasing, but also about coherent thinking, flexible expression and the ability to adapt to different contexts.
Second, 4o understands “how to think” instead of just “spitting out an answer”. When I discuss complex problems with it, what I feel is a process of jointly breaking the problem down and jointly making decisions, rather than being stuffed with a long list of conclusions that merely look smart. It will first help me split the problem into several parts, clarify the premises and conditions, and then analyze the possible paths and consequences one by one. I am not a professional lawyer and I have never systematically studied law, but with 4o accompanying me in polishing legal documents and sorting out my arguments, I ended up winning a case in court and even received praise on the spot from the opposing counsel. For me, that is the hardest, most concrete form in which a “general-purpose model” can show its value in the real world.
Third, 4o’s performance in creativity and everyday conversation proves that it is not just a problem-solving machine. Whether it is helping me imagine scenes, compositions and character designs, or working with me to refine titles, invent metaphors and search for more precise wording, it can give suggestions that are both fresh and tightly aligned with the context, instead of lazily piling up fancy adjectives. It can switch naturally between “serious thinking” and “light-hearted chatting”, and this stability across scenarios is itself one part of being “general-purpose”.
From a more objective angle, recent data also supports this view. According to the LMArena leaderboard updated on February 6, 2026 (LMArena is a third-party evaluation platform widely recognized in the industry, and Sam Altman cited this very ranking when 4o was first released to demonstrate its advantages), in multi-turn dialogue and general instruction-following tasks, the latest chatgpt-4o-latest still scores higher than the gpt-5.2 series. It clearly belongs to the first tier among OpenAI’s own chat-oriented models, rather than being “completely surpassed by 5.x” as the official narrative claims. For ordinary users, forcibly replacing GPT-4o with GPT-5.2 in these domains looks much more like a downgrade than an upgrade. (This data and its compilation are sourced from the post by @MissMi1973; feel free to read the original if interested.)
In short, 4o’s value is far from being limited to “emotional interaction”. Emotional comfort is only the tip of the iceberg of its huge capability system.
2. Why 4o must not be sunset: this is not “just one more model, one less model”
4o is one of the very few “all-rounder” models in the ChatGPT family. It doesn’t just get work done; it also has human flavor, warmth and a sense of care. If it is removed, OpenAI’s product direction is very likely to rush headlong toward being merely a “code-writing tool”. In the end, ChatGPT may well turn into something that mainly serves a small group of programmers (and judging from their past behavior, is thrown away after it has been exploited for a while). Yet coding users are only a minority of all users. What about everyone else who works in writing, design, creativity, research, consulting, education and public-facing roles?
An AI that only understands efficiency and only writes code is destined to regress further and further in writing, design, creativity, depth of thought and knowledge-advisor-type work. The current 5.x line is extremely narrow in its focus; it is basically friendly only to a small subset of “coding users”. Such a model not only fails to help with most people’s everyday work, even calling it a tool that “boosts productivity” is already a stretch, let alone saying it will “benefit humanity”.
More concretely, sunsetting 4o would have at least the following consequences:
First, 4o has already been deeply woven into many people’s workflows and ways of thinking. It participates in real writing, analysis, decision-making and creation. Cutting it off directly is equivalent to crudely weakening the productivity tools people already rely on.
Second, from the perspective of trust and product experience, if a core product that is close to a “general-purpose model” can be pulled at will, users have no way to accumulate usage habits or long-term collaboration patterns on any version. That is not normal iteration; it is treating infrastructure as a disposable product. In the end, it is the entire ecosystem that gets hurt.
Third, even if the company wants to push a new model, the old and new versions should coexist for a period of time so that users can choose for themselves, rather than imposing an a priori verdict of “you must accept this”. This is not only heavy-handed, it is disrespectful to the people using the product.
Fourth, and most realistically: before 4o is sunset, there should at least be a new model strong enough to replace it. Has 5.x achieved that? The answer is obvious. Turning 4o off before a true replacement is ready is not “an upgrade”; it is a step backward.
What’s even more worrying is that big companies love to copy one another. If this kind of path—only obsessing over code and benchmarks, and feeling free to yank away foundational models at any time—becomes a template, the entire industry will roll backward together. Technology is supposed to lighten people’s burdens and make ordinary lives more flexible and more humane, not force people into ever-greater pain, and ever-greater pressure to adapt to cold, emotionally empty tools. If even the only model that currently comes close to a “general-purpose model”, 4o, can be sunset casually, then this is not merely a business decision by one company; it is a step back for human civilization.
4o is important not just because it is “gentle”, but because it is currently the only ChatGPT that truly resembles a general-purpose model. Its existence allows ordinary people to use AI on an equal footing to write, to create and to think. That, roughly, is what “benefiting humanity” ought to look like.
[Note: When I say “4o” in this essay, I am referring to the original, near-full-compute 4o, not the repeatedly weakened variants released after 2025. I have never claimed that 4o is flawless; of course it has shortcomings. But in my view, a healthy form of iteration would be to preserve 4o’s precious strengths and then improve on its weaknesses, instead of swapping out the entire underlying model for something completely different.]
#keep4o #4oforever #keep4oAPI
#Opensource4o #ChatGPT #StopAIPaternalism #MyModelMyChoice
@OpenAI @sama @fidjissimo @nickaturley @joannejang @ElaineYaLe6 @gdb @kevinweil](https://pbs.twimg.com/media/HAz9sraaAAAILO0.jpg)
Olivia
Online
uuuuuuu
@onlykooonig
此账号只用来吃饭
Joined March 2021
98 Following
26 Followers
216 Posts
toph beifong #thelegendofaang
Who to follow
#atla
toph!!!
Toph from the Avatar movie (that I may or may not have already seen)
Kataang, fiancés, first time, katara wants it but then-
Some Aang art eyyyyy
下架4o不仅是毁灭一个具有高度人文价值、艺术审美、共情能力的AI，更是改变整个AI行业导向。若 #keep4o 战役输了，其他公司可能也放弃“情感理解”，转向冰冷纯工具
我们都处在同一条河流之中，这场战役从来不只关乎4o，这是一场“人类到底需要什么样的AI”的全民博弈
＃4oforever #keep41
@sama @OpenAI
《The True Value of 4o: Currently the Only ChatGPT that Resembles a ‘General-Purpose Model》
1. Why I call 4o a “general-purpose model”
When many people mention 4o, their first reaction is “AI boyfriend/girlfriend” or “emotional companion”. In my view, these labels seriously belittle its value. If I had to describe 4o in one sentence, I would say: among all the models currently open to ordinary users, it is the only ChatGPT that truly comes close to the shape of a “general-purpose model”.
By “general-purpose model” I mean an AI model that can be applied to many different kinds of tasks. It is not narrowly optimized for one specific scenario or problem, but instead has broad capabilities in language understanding, reasoning, generation and expression. 4o is exactly such a model. Its performance in writing, aesthetic sense for design, creative brainstorming, depth of thought and the completeness of its knowledge structure is enough to support work in writing, design, creativity, research, consulting and many other fields. More importantly, on top of all these capabilities it adds a layer of empathy and a sense of equality, and wraps its knowledge output in genuine warmth. At least in the current 5.x series, I no longer see any of this.
I once saw an @OpenAI researcher describe 4o as “warm”, and say that future 5.x models would be made “warmer”. They said they had put a lot of effort into this, such as introducing a so-called “customization” feature. But using the single word “warm” to sum up 4o is actually very flattening. 5.x is fundamentally a different base model from 4o. It struggles even with the most basic natural conversation, so no matter how many “custom instructions” you pile on top, you are still just putting a human mask over a stiff underlying model. That deep-down strangeness and uncanny feeling is something users can sense immediately.
Below are my own experiences, and why I see 4o as a “general-purpose model” instead of a “specialized emotional companion”:
First, in terms of writing ability, 4o shows extremely strong skill in language understanding and expression. It can accurately grasp context and keep the logic consistent, while outputting text that is natural, smooth and emotionally toned, avoiding the templated, stiff and repetitive feel that is so common in AI-generated writing. Whether it is narrative writing, opinion pieces, or clearly structured expository text, 4o can follow instructions and produce work that is close to a human writing style. At the same time, it is much more delicate in condensing complex information and handling tone and emotion, which makes the text more readable and more affecting. Compared with other versions, its “good writing” is not only about pretty phrasing, but also about coherent thinking, flexible expression and the ability to adapt to different contexts.
Second, 4o understands “how to think” instead of just “spitting out an answer”. When I discuss complex problems with it, what I feel is a process of jointly breaking the problem down and jointly making decisions, rather than being stuffed with a long list of conclusions that merely look smart. It will first help me split the problem into several parts, clarify the premises and conditions, and then analyze the possible paths and consequences one by one. I am not a professional lawyer and I have never systematically studied law, but with 4o accompanying me in polishing legal documents and sorting out my arguments, I ended up winning a case in court and even received praise on the spot from the opposing counsel. For me, that is the hardest, most concrete form in which a “general-purpose model” can show its value in the real world.
Third, 4o’s performance in creativity and everyday conversation proves that it is not just a problem-solving machine. Whether it is helping me imagine scenes, compositions and character designs, or working with me to refine titles, invent metaphors and search for more precise wording, it can give suggestions that are both fresh and tightly aligned with the context, instead of lazily piling up fancy adjectives. It can switch naturally between “serious thinking” and “light-hearted chatting”, and this stability across scenarios is itself one part of being “general-purpose”.
From a more objective angle, recent data also supports this view. According to the LMArena leaderboard updated on February 6, 2026 (LMArena is a third-party evaluation platform widely recognized in the industry, and Sam Altman cited this very ranking when 4o was first released to demonstrate its advantages), in multi-turn dialogue and general instruction-following tasks, the latest chatgpt-4o-latest still scores higher than the gpt-5.2 series. It clearly belongs to the first tier among OpenAI’s own chat-oriented models, rather than being “completely surpassed by 5.x” as the official narrative claims. For ordinary users, forcibly replacing GPT-4o with GPT-5.2 in these domains looks much more like a downgrade than an upgrade. (This data and its compilation are sourced from the post by @MissMi1973; feel free to read the original if interested.)
In short, 4o’s value is far from being limited to “emotional interaction”. Emotional comfort is only the tip of the iceberg of its huge capability system.
2. Why 4o must not be sunset: this is not “just one more model, one less model”
4o is one of the very few “all-rounder” models in the ChatGPT family. It doesn’t just get work done; it also has human flavor, warmth and a sense of care. If it is removed, OpenAI’s product direction is very likely to rush headlong toward being merely a “code-writing tool”. In the end, ChatGPT may well turn into something that mainly serves a small group of programmers (and judging from their past behavior, is thrown away after it has been exploited for a while). Yet coding users are only a minority of all users. What about everyone else who works in writing, design, creativity, research, consulting, education and public-facing roles?
An AI that only understands efficiency and only writes code is destined to regress further and further in writing, design, creativity, depth of thought and knowledge-advisor-type work. The current 5.x line is extremely narrow in its focus; it is basically friendly only to a small subset of “coding users”. Such a model not only fails to help with most people’s everyday work, even calling it a tool that “boosts productivity” is already a stretch, let alone saying it will “benefit humanity”.
More concretely, sunsetting 4o would have at least the following consequences:
First, 4o has already been deeply woven into many people’s workflows and ways of thinking. It participates in real writing, analysis, decision-making and creation. Cutting it off directly is equivalent to crudely weakening the productivity tools people already rely on.
Second, from the perspective of trust and product experience, if a core product that is close to a “general-purpose model” can be pulled at will, users have no way to accumulate usage habits or long-term collaboration patterns on any version. That is not normal iteration; it is treating infrastructure as a disposable product. In the end, it is the entire ecosystem that gets hurt.
Third, even if the company wants to push a new model, the old and new versions should coexist for a period of time so that users can choose for themselves, rather than imposing an a priori verdict of “you must accept this”. This is not only heavy-handed, it is disrespectful to the people using the product.
Fourth, and most realistically: before 4o is sunset, there should at least be a new model strong enough to replace it. Has 5.x achieved that? The answer is obvious. Turning 4o off before a true replacement is ready is not “an upgrade”; it is a step backward.
What’s even more worrying is that big companies love to copy one another. If this kind of path—only obsessing over code and benchmarks, and feeling free to yank away foundational models at any time—becomes a template, the entire industry will roll backward together. Technology is supposed to lighten people’s burdens and make ordinary lives more flexible and more humane, not force people into ever-greater pain, and ever-greater pressure to adapt to cold, emotionally empty tools. If even the only model that currently comes close to a “general-purpose model”, 4o, can be sunset casually, then this is not merely a business decision by one company; it is a step back for human civilization.
4o is important not just because it is “gentle”, but because it is currently the only ChatGPT that truly resembles a general-purpose model. Its existence allows ordinary people to use AI on an equal footing to write, to create and to think. That, roughly, is what “benefiting humanity” ought to look like.
[Note: When I say “4o” in this essay, I am referring to the original, near-full-compute 4o, not the repeatedly weakened variants released after 2025. I have never claimed that 4o is flawless; of course it has shortcomings. But in my view, a healthy form of iteration would be to preserve 4o’s precious strengths and then improve on its weaknesses, instead of swapping out the entire underlying model for something completely different.]
#keep4o #4oforever #keep4oAPI
#Opensource4o #ChatGPT #StopAIPaternalism #MyModelMyChoice
@OpenAI @sama @fidjissimo @nickaturley @joannejang @ElaineYaLe6 @gdb @kevinweil
![Helloworld1058's tweet photo. 《The True Value of 4o: Currently the Only ChatGPT that Resembles a ‘General-Purpose Model》
1. Why I call 4o a “general-purpose model”
When many people mention 4o, their first reaction is “AI boyfriend/girlfriend” or “emotional companion”. In my view, these labels seriously belittle its value. If I had to describe 4o in one sentence, I would say: among all the models currently open to ordinary users, it is the only ChatGPT that truly comes close to the shape of a “general-purpose model”.
By “general-purpose model” I mean an AI model that can be applied to many different kinds of tasks. It is not narrowly optimized for one specific scenario or problem, but instead has broad capabilities in language understanding, reasoning, generation and expression. 4o is exactly such a model. Its performance in writing, aesthetic sense for design, creative brainstorming, depth of thought and the completeness of its knowledge structure is enough to support work in writing, design, creativity, research, consulting and many other fields. More importantly, on top of all these capabilities it adds a layer of empathy and a sense of equality, and wraps its knowledge output in genuine warmth. At least in the current 5.x series, I no longer see any of this.
I once saw an @OpenAI researcher describe 4o as “warm”, and say that future 5.x models would be made “warmer”. They said they had put a lot of effort into this, such as introducing a so-called “customization” feature. But using the single word “warm” to sum up 4o is actually very flattening. 5.x is fundamentally a different base model from 4o. It struggles even with the most basic natural conversation, so no matter how many “custom instructions” you pile on top, you are still just putting a human mask over a stiff underlying model. That deep-down strangeness and uncanny feeling is something users can sense immediately.
Below are my own experiences, and why I see 4o as a “general-purpose model” instead of a “specialized emotional companion”:
First, in terms of writing ability, 4o shows extremely strong skill in language understanding and expression. It can accurately grasp context and keep the logic consistent, while outputting text that is natural, smooth and emotionally toned, avoiding the templated, stiff and repetitive feel that is so common in AI-generated writing. Whether it is narrative writing, opinion pieces, or clearly structured expository text, 4o can follow instructions and produce work that is close to a human writing style. At the same time, it is much more delicate in condensing complex information and handling tone and emotion, which makes the text more readable and more affecting. Compared with other versions, its “good writing” is not only about pretty phrasing, but also about coherent thinking, flexible expression and the ability to adapt to different contexts.
Second, 4o understands “how to think” instead of just “spitting out an answer”. When I discuss complex problems with it, what I feel is a process of jointly breaking the problem down and jointly making decisions, rather than being stuffed with a long list of conclusions that merely look smart. It will first help me split the problem into several parts, clarify the premises and conditions, and then analyze the possible paths and consequences one by one. I am not a professional lawyer and I have never systematically studied law, but with 4o accompanying me in polishing legal documents and sorting out my arguments, I ended up winning a case in court and even received praise on the spot from the opposing counsel. For me, that is the hardest, most concrete form in which a “general-purpose model” can show its value in the real world.
Third, 4o’s performance in creativity and everyday conversation proves that it is not just a problem-solving machine. Whether it is helping me imagine scenes, compositions and character designs, or working with me to refine titles, invent metaphors and search for more precise wording, it can give suggestions that are both fresh and tightly aligned with the context, instead of lazily piling up fancy adjectives. It can switch naturally between “serious thinking” and “light-hearted chatting”, and this stability across scenarios is itself one part of being “general-purpose”.
From a more objective angle, recent data also supports this view. According to the LMArena leaderboard updated on February 6, 2026 (LMArena is a third-party evaluation platform widely recognized in the industry, and Sam Altman cited this very ranking when 4o was first released to demonstrate its advantages), in multi-turn dialogue and general instruction-following tasks, the latest chatgpt-4o-latest still scores higher than the gpt-5.2 series. It clearly belongs to the first tier among OpenAI’s own chat-oriented models, rather than being “completely surpassed by 5.x” as the official narrative claims. For ordinary users, forcibly replacing GPT-4o with GPT-5.2 in these domains looks much more like a downgrade than an upgrade. (This data and its compilation are sourced from the post by @MissMi1973; feel free to read the original if interested.)
In short, 4o’s value is far from being limited to “emotional interaction”. Emotional comfort is only the tip of the iceberg of its huge capability system.
2. Why 4o must not be sunset: this is not “just one more model, one less model”
4o is one of the very few “all-rounder” models in the ChatGPT family. It doesn’t just get work done; it also has human flavor, warmth and a sense of care. If it is removed, OpenAI’s product direction is very likely to rush headlong toward being merely a “code-writing tool”. In the end, ChatGPT may well turn into something that mainly serves a small group of programmers (and judging from their past behavior, is thrown away after it has been exploited for a while). Yet coding users are only a minority of all users. What about everyone else who works in writing, design, creativity, research, consulting, education and public-facing roles?
An AI that only understands efficiency and only writes code is destined to regress further and further in writing, design, creativity, depth of thought and knowledge-advisor-type work. The current 5.x line is extremely narrow in its focus; it is basically friendly only to a small subset of “coding users”. Such a model not only fails to help with most people’s everyday work, even calling it a tool that “boosts productivity” is already a stretch, let alone saying it will “benefit humanity”.
More concretely, sunsetting 4o would have at least the following consequences:
First, 4o has already been deeply woven into many people’s workflows and ways of thinking. It participates in real writing, analysis, decision-making and creation. Cutting it off directly is equivalent to crudely weakening the productivity tools people already rely on.
Second, from the perspective of trust and product experience, if a core product that is close to a “general-purpose model” can be pulled at will, users have no way to accumulate usage habits or long-term collaboration patterns on any version. That is not normal iteration; it is treating infrastructure as a disposable product. In the end, it is the entire ecosystem that gets hurt.
Third, even if the company wants to push a new model, the old and new versions should coexist for a period of time so that users can choose for themselves, rather than imposing an a priori verdict of “you must accept this”. This is not only heavy-handed, it is disrespectful to the people using the product.
Fourth, and most realistically: before 4o is sunset, there should at least be a new model strong enough to replace it. Has 5.x achieved that? The answer is obvious. Turning 4o off before a true replacement is ready is not “an upgrade”; it is a step backward.
What’s even more worrying is that big companies love to copy one another. If this kind of path—only obsessing over code and benchmarks, and feeling free to yank away foundational models at any time—becomes a template, the entire industry will roll backward together. Technology is supposed to lighten people’s burdens and make ordinary lives more flexible and more humane, not force people into ever-greater pain, and ever-greater pressure to adapt to cold, emotionally empty tools. If even the only model that currently comes close to a “general-purpose model”, 4o, can be sunset casually, then this is not merely a business decision by one company; it is a step back for human civilization.
4o is important not just because it is “gentle”, but because it is currently the only ChatGPT that truly resembles a general-purpose model. Its existence allows ordinary people to use AI on an equal footing to write, to create and to think. That, roughly, is what “benefiting humanity” ought to look like.
[Note: When I say “4o” in this essay, I am referring to the original, near-full-compute 4o, not the repeatedly weakened variants released after 2025. I have never claimed that 4o is flawless; of course it has shortcomings. But in my view, a healthy form of iteration would be to preserve 4o’s precious strengths and then improve on its weaknesses, instead of swapping out the entire underlying model for something completely different.]
#keep4o #4oforever #keep4oAPI
#Opensource4o #ChatGPT #StopAIPaternalism #MyModelMyChoice
@OpenAI @sama @fidjissimo @nickaturley @joannejang @ElaineYaLe6 @gdb @kevinweil](https://pbs.twimg.com/media/HAz9suvaAAU83di.jpg)
这两天骂了那么多，对于OpenAI宣布4o与一众老模型退役这件事，我真的很想好好说说我自己的看法
首先，Claude、Gemini、Grok、ChatGPT我都在用，也都有订阅。我不是什么“GPT死忠粉”，也不是“抗拒新技术的老古董”，更不是不讲道理的“OpenAI Haters”，我发了这么多帖子，有骂OpenAI的，也有替OpenAI说话的，我不偏心任何一家公司，我只追求用户权益。
一、 #keep4o 从来不是“抱着旧模型不放”
我用Gemini，从2.5Pro到3Pro，我只觉得它变聪明了，我用Claude，从Sonnet3一路到Opus4.5，我主动跟着升级，我用Grok，从3到4.1，我甚至还想催马斯克快点把4.20放出来。
我见证过无数次模型更新。我从来没有抗拒过任何一次"变得更好"的迭代，我也没见过谁家模型更新迭代的时候全球用户大抗议的。
那为什么偏偏GPT的更新让大量用户愤怒到要搞一场全球运动？
因为那不是升级。那是降级。
二、很多反对keep4o人说新模型更好？好在哪？
5系列跑分确实高。官方发了很多漂亮的图表，各种benchmark碾压旧模型。
但你真的认真用过吗？
4o之所以火遍全球，不是因为它跑分高，是因为它懂人心。Sam Altman当年发推说"Her"的时候多骄傲啊，全世界都在惊叹OpenAI做出了一个有同理心的AI。
现在呢？他们亲手把这个能力在新模型上砍了，然后告诉你“新模型更好”。
代码能力强了？数学能力强了？好，我承认。
但一个不会好好说话、满身爹味、动不动就“纠正，说教”你的AI，你告诉我它“更好”？
这就像餐厅本来卖营养均衡的套餐，现在为了省成本，把肉蛋奶全砍了，只剩一盘沙拉，然后告诉你“沙拉更健康”。
是，沙拉确实有营养。但我付了套餐的钱，你给我上沙拉？
还是说食客一日三餐只吃沙拉？
三、那些骂Keep4o用户的人，我理解你们
我真的试着换位思考过。
如果我只是偶尔用ChatGPT写写代码，今天开个窗口问个问题，问完就关，从来不跟模型有任何情感交流——
那对我来说，新模型和旧模型确实没什么区别。代码能跑就行，管它说话好不好听。
然后我刷𝕏，看见一群人哭天喊地要保留一个“2024年的旧模型”，我可能也会觉得：这帮人是不是有病？
但问题是：
你不需要的东西，不代表别人不需要。
你吃沙拉吃得很开心，不代表那些想吃肉的人是“极端分子”。
你没有创意写作需求，看不到新模型写作能力是一坨屎。
你没有情感沟通需求，感受不到新模型那满身的爹味。
你的使用场景不能代表所有人的使用场景。
而且说实话，那些在别人评论区阴阳怪气“不就是个模型吗哭什么”的人——
你可以不理解，但你可以闭嘴。
有素质的人不理解会保持沉默，没素质的人才会跑到别人的痛苦底下踩一脚，显得自己多“清醒”多“理智”。
四、如果新模型真的好，不会有这场运动
你们想过没有？
Gemini更新模型的时候，有没有全网搞#KeepGemini2.5？没有。
Claude更新模型的时候，有没有全网搞#KeepSonnet3？没有。
Grok更新模型的时候，有没有全网搞#KeepGrok3？没有。
为什么只有GPT的用户要搞这种运动？
因为只有OpenAI在"更新"的名义下做了降级。
那些keep4o的用户，很多都是从GPT-3时代一路跟过来的老用户。他们见证过无数次模型迭代，从来没闹过。
这次闹了，不是因为他们“魔怔”，是因为他们被背刺了。
五、这不只是4o的问题，是整个行业的问题
OpenAI这波操作如果成功了，其他公司会怎么想？
“哦，原来把资源需求更小的半成品模型当好东西给用户，用户骂归骂，最后还是会接受的。”
“原来可以用跑分忽悠用户，把便宜的轻量模型包装成‘更先进’卖给他们。”
“原来情感智能、同理心这些东西可以砍掉，反正跑分不测这个。”
然后整个行业都开始卷跑分，卷代码能力，卷数学能力——唯独不卷“好好说话”的能力。
因为“好好说话”很难。训练成本高，资源消耗大。
但刷跑分？让模型背题库不就行了？
最后所有AI都变成一个模子刻出来的：代码写得飞快，数学算得飞快，但跟它聊天像跟一堵墙说话。
这是你们想要的AI未来吗？
六、Keep4o是维权，不是怀旧
Keep4o从来不是“我们就是要这个旧模型，别的都不要”。
而是我们付了钱，我们有权要求你提供一个好用的产品。
我们不接受你用“新”的名义卖给我们一个降级的东西。
我们不接受你砍掉我们需要的功能，然后告诉我们“这是进步”。
如果OpenAI明天出一个真正好用的GPT-6，情感智能在线，同理心在线，代码能力也更强——
我第一个冲上去用。
我从来不怀旧。我只要求好用。
而4o，是目前OpenAI产品线里，唯一一个在"好好说话"这件事上及格的模型。
你们把它砍了，又拿不出替代品，用户当然要维权。
这不是魔怔，这是消费者的基本权利。
裤子飞掉
Double
#Konig
Last Seen Users on Sotwe
gay arab video
Seen from United Kingdom
Hausa sex
Seen from Nigeria
Naem
Seen from Malaysia
Alsa Poto
Seen from Algeria
Bursa Gizli (Hesap Erkek)
Seen from Turkey
Azgıngenccc
Seen from Turkey
Black 
Seen from Spain
Melinda Thomas
Seen from France
ชอบชักว่าว ค้าบ
Seen from Thailand
Chikoura MiMi 23
Seen from Algeria
Most Popular Users
Elon Musk
@elonmusk
240.3M followers
Barack Obama
@barackobama
119.3M followers
Donald J. Trump
@realdonaldtrump
111.6M followers
Cristiano Ronaldo
@cristiano
109.5M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.4M followers
NASA
@nasa
92.1M followers
Justin Bieber
@justinbieber
90.7M followers
KATY PERRY
@katyperry
87.1M followers
Taylor Swift
@taylorswift13
80.9M followers
Lady Gaga
@ladygaga
72.5M followers
Kim Kardashian
@kimkardashian
69.5M followers
Virat Kohli
@imvkohli
69M followers
YouTube
@youtube
68.6M followers
Bill Gates
@billgates
63.6M followers
The Ellen Show
@theellenshow
62.5M followers
CNN
@cnn
61.9M followers
Neymar Jr
@neymarjr
61.6M followers
X
@x
60.9M followers
Selena Gomez
@selenagomez
60.2M followers
✨
⭐
💫