GPT-4 officially retires! The most powerful model ever, officially handed over to GPT-4.1, o3, and o4 mini!

The AI ​​circle in April is destined to be turbulent.

On April 10, OpenAI issued an announcement to send away many of ChatGPT Plus users:

GPT-4 It will be officially decommissioned from ChatGPT on April 30 and replaced by a more powerful GPT-4o Comprehensive replacement.

Official words:

“GPT-4o "Overall surpassed" in writing, programming, STEM, etc. GPT-4", is already enough to assume the role of "successor."

Attached here is the OpenAI announcement in both Chinese and English, to commemorate the team that once “blasted” the world and stayed with us for two years. GPT-4.

GPT-4, which was once a "milestone" in the entire AI world.

It is the first large language model (LLM) that can truly "write articles, compile code, and understand complex logic". It is the watershed that transformed ChatGPT from a toy into a tool, and it is also the moment when countless people first felt the reality of AGI (artificial general intelligence).

It is OpenAI's trump card to expand its global user base and the first window for AI tools to enter the lives of ordinary people.

In the words of netizens:

"No GPT-4, we won’t feel that the AI ​​era has really arrived so early.”

However, it all comes to an end on April 30th.

ChatGPT GPT-4(Actually, more precisely,GPT-4 Turbo) The model's introduction has also been updated.

exist GPT-4 Before it retires, let us take a look at its splendor in its old age.

The one we are familiar with GPT-4, it's leaving.

So, our current situation is...

GPT-4 retired;

GPT-4o took over;

But the story is far from over.

Because just GPT-4 While he is about to retire, OpenAI has already been preparing a large number of new players.

GPT-4.1,GPT-4.1 mini/nano,o3,o4 mini.

New models with weirder names are lining up to enter the market.

A "spring refresh" of the GPT model is quietly underway.

And what's coming up next is a whole GPT-4.1 + o Series family.

According to The Verge and other solid news on the Internet, the new OpenAI models that can be foreseen include:

1. GPT-4.1: An upgraded version of GPT-4o

GPT-4.1 is GPT-4o(can process text, images, and voice) based on the iteration, the internal positioning is similar GPT-4o plus, are rumored to have faster response times, better visual understanding, and more natural audio input interactions.

Posted over a month ago GPT-4.5 It was undoubtedly a failure, mainly because the parameters were too large and the cost was too high, and OpenAI simply could not roll it out on a large scale for users to use.

To combine the "non-thinking" model with the reasoning model GPT-5 It is said that various difficulties were encountered again.

in this case,GPT-4.1 It may be GPT-4o A transitional version of .

The official has not yet disclosed the specific features, but outsiders speculate:

• Faster response time
• Stronger multimodal (image, audio) capabilities
• Long context/dialogue memory is more stable

"A streamlined, efficient, more modern version" GPT-4.1 It may be officially released next week.

2. GPT-4.1 mini/nano: lightweight models

GPT-4.1 A mini version of GPT-4o and GPT-4o mini relationship.

GPT-4.1 mini/nano This type of model focuses on "low latency and low resource consumption" and is suitable for deployment on edge devices (such as mobile apps, applets, and embedded scenarios). It is closer to the positioning of small parameters, low fuel consumption, and cost-effectiveness.

The simple understanding is: "It may not be the strongest, but you can use it at any time."

3. o3, o4 mini: next generation inference models

o The series is an independent naming system outside the GPT family, representing the new generation of "inference model structure" incubated by OpenAI.

• o3 is a "reasoning first" architecture (stronger complex instruction execution capabilities)
• o4 mini may be a more powerful but low-computing-power-friendly compression model

Netizens have already found references to these models in the new version of ChatGPT page, which is enough to prove that the countdown to the launch has begun.

Despite having so many new models, OpenAI faces a problem: not enough GPU/computing power.

Sam Altman has admitted many times on X that the biggest problem at the moment is not the model, but "the graphics card supply cannot keep up."

The original words are very straightforward:

“Our GPUs are about to burn out.”

"We're dealing with capacity challenges, and sometimes service will be slow, and some things may be delayed."

This also explains why:

• Sora is temporarily closed to new users
• Image generation functions were previously limited in frequency
• The new model has not yet released full functionality
• GPT-5 has been “trained for a long time”, but it has not yet been launched

OpenAI seems to be trying to strike a balance between iterating new models, optimizing hardware resources, and keeping the product stable.

Finally, let’s briefly review OpenAI’s current pace.

Conclusion: The exit of GPT-4, the preparation for GPT-5?

Sam Altman once suggested GPT-5 Training is ongoing, but “not that fast” because he hopes this generation of models will achieve the goal of being more stable, more consistent, and smarter.

To put it bluntly, they haven't been trained well enough yet.

So currently these GPT-4.1,o3,o4 miniThis is OpenAI’s “transition strategy” this quarter, which is to handle traffic while optimizing the architecture to see which line can run first.

and GPT-4 The final chapter of the 21st Century Business Herald is the prologue to our deep integration with AI. The true “next generation” has just begun.