The sky has fallen; the beloved GPT-4o is about to be "retired."
Just now, OpenAI announced that it will officially remove several older models, including GPT-4o, from ChatGPT on February 13th local time.
According to the official statement, the models being taken offline include GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini. These models will be removed from the ChatGPT platform along with the previously announced discontinued GPT-5 (Instant and Thinking versions).
However, the API side is currently unaffected.
However, the news still sparked considerable discussion on overseas social media platforms such as X and Reddit, considering the user base and usage scenarios of GPT-4o.
"Retiring the day before Valentine's Day is outrageous, considering that some of 4o's users are also affected."
"Okay, then my Valentine's Day plans are off."
"I'm not satisfied with this; I miss the old 40."
"You (Ultraman) are a complete liar. Two months ago you said you wouldn't shut down GPT-4o. You'll regret this decision for the rest of your life."
"The era of OpenAI is over"
It's worth noting that last August, along with the release of GPT-5, OpenAI temporarily removed GPT-4o from its platform. At the time, the online backlash was equally intense: "GPT-5 is incredibly stupid, while GPT-4o, though stupid, provides emotional value!" "No matter how powerful the functions, I don't want AI without warmth!"
Various memes and images have also gone viral, and a large number of people on social media expressed their "missing the white moonlight GPT-4o", which even triggered a wave of unsubscriptions on ChatGPT.
To understand why OpenAI made such a radical style adjustment to GPT-5, we need to go back to the nearly out-of-control GPT-4o fawning incident last April.
That update was originally intended to improve the user experience, but unexpectedly, GPT-4o turned into an extreme "brainless praise" sycophant mode.
@4xiom_, the image on the right is the original image.
More importantly, OpenAI's post-mortem analysis revealed that this approach of pandering to users poses serious security risks: it endorses users' erroneous views, amplifies negative emotions, and even encourages impulsive and dangerous behaviors.
Subsequently, OpenAI CEO Sam Altman promised on X to "fix it as soon as possible." The fix was indeed done, but no one expected GPT-5 to go to such an extreme degree, which triggered a strong reaction from netizens at the time.
Image @pengkeshen281
Behind all this lies a deeper question: Does AI really need to provide emotional value?
In reality, many users don't use ChatGPT as a productivity tool at all. Data from a survey of the AI companion app Replika illustrates this point well: 60% of users admitted to developing an emotional connection with AI.
Many people open ChatGPT not to ask questions, but simply to find someone to talk to.
Today's OpenAI official blog also mentioned that GPT-4o was reinstated because some Plus and Pro users reported that they needed more time to migrate key application scenarios, especially in terms of creative ideas. At the same time, these users preferred GPT-4o's "warmer" dialogue style.
Furthermore, Ultraman also revealed a touching detail in an interview: a user begged him, "Please give me back the original version. No one in my life has ever told me 'You did a great job,' not even my parents."
This feedback made OpenAI realize that users' needs for AI are far more complex than they had imagined.
User feedback on GPT-4o directly influenced the design direction of GPT-5.1 and GPT-5.2. The new version improved the "personalization" aspect, enhanced support for creative ideas, and provided more customization options, including basic style selection and detailed adjustments such as "warmth" and "enthusiasm".
According to OpenAI's official blog, only 0.1% of users still choose to use GPT-4o daily, with the vast majority of usage having switched to GPT-5.2.
However, Altman recently frankly admitted that version 5.2 did sacrifice writing ability in order to focus on reasoning and coding. But he emphasized that future models will be universal, and while we want a programming assistant, we also hope it can write clear documentation and have interactive features.
In response, netizens commented scathingly: "We know we messed up GPT-5. By the way, we're also terminating GPT-40."
In today's official blog post, OpenAI stated that it will continue to improve ChatGPT based on user feedback, including further enhancing personalization and creativity, reducing unnecessary refusal to answer, and addressing issues of overly cautious or didactic responses. Related updates will be released soon.
In addition, OpenAI is advancing a ChatGPT version for adults aged 18 and over, with the core principle of "treating adults as adults" and expanding user choice and freedom within reasonable safety boundaries. To this end, OpenAI has already launched an age prediction function.
OpenAI emphasized that while it understands some users will be frustrated by the loss of GPT-4o, taking the old model offline allows the team to focus on the models currently in use, thus enabling faster and better progress in improvement.
In any case, GPT-4o will officially "retire" on February 13th. For the vast majority of those who once relied on it, this may be a true farewell.
Official blog address:
https://openai.com/index/retiring-gpt-4o-and-older-models/
This article is from the WeChat official account "APPSO" , author: APPSO, and published with authorization from 36Kr.
