Key Points:

1. After two years of anticipation, GPT-5 has arrived, surpassing the smartest large models globally and reaching a doctoral-level intelligence.

2. GPT-5 offers free, plus, and Pro modes for general users, with GPT-5, GPT-5 nano, and GPT-5 mini models available on the API platform.

3. A significant error occurred during OpenAI's live broadcast, causing widespread criticism from netizens over the displayed performance chart.

Author: Chang Yuan

Editor: Key Points Master

GPT-5 was finally unveiled in the wee hours of the night.

Were you shocked? Were you amazed? Indeed, many were.

But first, take a look at this image:

How did OpenAI manage to display a Benchmark during a global live broadcast where 52.8 was shown higher than 69.1, and even this 69.1 was aligned with 30.8???

Setting aside other issues, this one point alone, where the AI, initially heralded as "doctoral-level," cleverly "scaled the coordinate system as needed," has sparked widespread criticism from netizens.

Even Altman quickly came out to change the subject, stating that the technical blog was correct...

Indeed, the technical blog did make corrections.

However, such a mistake is truly unacceptable, especially after two years of anticipation! An AI of doctoral-level intelligence has finally arrived.

Despite this fatal error that has been widely criticized, GPT-5's on-stage performance was quite remarkable.

In the technical blog, OpenAI began by stating, "This is our most intelligent, fastest, and most practical model to date, with built-in thinking capabilities that allow everyone to have expert-level intelligence."

The GPT-5 released this time comes in four versions:

GPT-5: The standard mode for coding and executing tasks across various fields;

GPT-5 mini: A lightweight version suitable for clearly defined tasks and scenarios;

GPT-5 nano: Emphasizes running speed and cost-effectiveness;

GPT-5 Chat: The version used in ChatGPT.

GPT-5 currently offers free, plus, and Pro modes for general users.

Meanwhile, GPT-5, GPT-5 nano, and GPT-5 mini models are available on the API platform.

Moreover, following yesterday's open-source release after a six-year hiatus, OpenAI announced that GPT-5 is now available to everyone for free, boasting a doctoral-level intelligence.

Let's first take a look at the Benchmark.

The most eye-catching aspect is GPT-5's performance at AIME 2025, where it scored a perfect score.

Next is its programming ability. Compared to o3 and 4o, GPT-5, supported by its thinking mode, reached a level of 74.9%.

Additionally, this model performed exceptionally well in various multimodal evaluations, covering aspects such as image, video, spatial understanding, and scientific reasoning.

Stronger multimodal capabilities mean that ChatGPT is smarter when processing images and other non-textual information - for instance, understanding charts, summarizing the content of a presentation photo, or answering questions about illustrations.

Rarely, a third-party large model arena (Imarena.ai) promptly followed up and directly stated, "First in all aspects."

Specifically, the arena covers content including text, web development, vision, programming, mathematics, creativity, long queries, etc.

It is evident that, based on current evaluation standards, GPT-5 is indeed the reigning champion among AI large models.

Now let's examine the effects.

While achievements are important, actual effects are the hard truth.

Well-versed in this, Sam Altman immediately followed up on his X account and published the effects generated by GPT-5:

Altman also mentioned that users with GPT-5 access can simply send "use beatbot to make a sick beat to celebrate gpt-5" to experience it.

To say the least, Altman is a marketing master.

However, during the live demonstration, GPT-5 still showcased many impressive performances.

For example, asking GPT-5 to generate a grammar learning app produced this result:

Don't think it's just a simple website. Besides handling AI interactions available on the market, it can also embed a mini-game inside (click "Mouse&Cheese"):

And if you want to change the content inside the app, just say a word, and the layout can be altered instantly (as tested on-site):

For larger and more complex projects, such as the code programmed by GPT-5 demonstrated on-site:

After running it, a 3D "world" was obtained:

In summary, based on the on-site effects, GPT-5 can indeed be considered the "smartest and strongest programming" large model.

Are the "audiences" buying it?

Judging from the feedback of "audiences" on X currently, the most heated discussion revolves around the chart bug we mentioned earlier.

Some netizens even created a simple and brutal Excel table to satirize this mistake:

On the other hand, Microsoft's CEO, Altman's former employer, quickly came out to support it with a "heart" emoji:

Another example is VS Code, commonly used by developers, which officially and seamlessly integrated GPT-5 on Day 0:

But for the general public, the most common feedback is, "It was said to be free, but why don't I have it on my ChatGPT?"

In conclusion, whether users are embracing GPT-5 and whether the actual effects are as impressive as claimed will require more time to ascertain.

Reference Link: