OpenAI's Stunning Revelation: GPT-5 Has Truly "Lost Intelligence", but It Replicates the "Masterstroke" and Aims for the Throne in Coding

Did GPT-5 only score 70 points in the IQ test? Behind the widespread online criticism of its "diminished intelligence", the truth is that "routing" determines the model's intelligence. The secret to unlocking the god - level GPT-5 lies in the prompt. Indeed, medical experts have used GPT-5 to recreate a "divine move" moment.

72 hours after the release of GPT-5, an IQ test result shocked the entire internet.

In the Mensa IQ test, GPT-5 scored 118 points, and 70 points in the offline test; GPT-5 Thinking scored 85 points and 57 points respectively.

This result set the lowest record in the IQ tests of OpenAI's model family.

Actually, the real reason behind this is attributed to the "routing" problem.

It's not that GPT-5 is too stupid. As a "single - body model", one of its components determines its intelligence.

Similar issues were also addressed by Altman in a Reddit AMA Q&A.

He said that there was a serious internal failure (Sev level), and the automatic switching system failed to work, causing GPT-5 to appear to have diminished intelligence.

According to the latest report from METR, GPT-5 still remains at the Pareto frontier, and its intelligence growth in an exponential manner has not slowed down.

That is to say, GPT-5 is still continuing the myth of the Scaling Law.

GPT-5 is powerful, and the key lies in the prompt

Those netizens who blindly criticize GPT-5 actually haven't discovered the potential of the latest model.

Cline, the head of artificial intelligence, said that the core lies in a person's ideas, taste, and communication style.

For users with systematic thinking, GPT-5 is a revolutionary tool. As long as they are willing to spend time: build a complete thinking framework, formulate clear requirements, and clearly explain them to the model.

Then, it can execute accurately on its own without any manual correction throughout the process.

Coincidentally, Mark Manson, the author of a NYT best - selling book, also said that everyone is talking to GPT-5 in the wrong way, and the key is to take the initiative.

In this way, when it knows you're not easy to fool, it will give perfect answers.

For example, if you want to ask how many "b"s are in "blueberry" and threaten it with "watch out for Bambi's mom if you answer wrong".

At this time, GPT-5 won't make a mistake.

Another example is that netizens were arguing that GPT-5 couldn't solve a simple equation. The actual trick also lies in the prompt.

When the prompt becomes "think harder and solve", the correct solution can be obtained.

What kind of prompt is effective? A netizen exposed the system prompt of GPT-5, which is like a gold mine.

The "divine move" moment

In the medical field, GPT-5 can already rival human experts.

Biomedical expert Derya Unutmaz, after experiencing GPT-5, deeply felt the "37th move" moment of AlphaGo.

The situation is like this. Two years ago, Derya's laboratory conducted a series of cutting - edge immunology experiments aiming to regulate the energy metabolism of T cells.

This kind of immune cell has a significant impact on cancer immunotherapy, chronic diseases, and autoimmune diseases.

At that time, they obtained an amazing result, but there was a discovery that they couldn't explain.

The team spent several weeks on it and only got partial answers.

Based on these experiments, Derya uploaded the unpublished data graph to GPT-5 Pro for analysis, and the result was astonishing.

With just the above graph, GPT-5 accurately identified the key discovery and provided suggestions for the experimental scheme.

Most incredibly, the mechanism it proposed finally explained all the results.

Derya Unutmaz said that this is simply a "divine move" moment in the field of AI. This process proves that GPT-5 has become a top - notch expert and a real scientific research partner, capable of providing profound insights.

OpenAI aims GPT-5 at Anthropic's throne

Although GPT-5 is not AGI yet, its powerful programming ability has attracted more developers.

In addition, its new personalized options and the reduced "hallucination" phenomenon may attract more daily users to the free version of ChatGPT.

This is undoubtedly a challenge to Anthropic.

The reason is that the most powerful AI model for writing code is generally recognized as Anthropic's Claude model.

Therefore, when OpenAI released the new model, it strongly emphasized the powerful programming ability of GPT-5.

GPT-5 is the most powerful programming model we have ever had. It performs particularly well in complex front - end generation and debugging large - scale code libraries.

With just one prompt, it can intuitively and elegantly create beautiful, responsive websites, applications, and games, turning ideas into reality.

The intention is very obvious.

At the press conference, Altman said that the new model is not only good at coding but also can transform software projects from ideas directly into usable code.

Various programs generated by GPT-5

Pietro Schirano, the CEO of the AI startup MagicPath, called GPT-5 the best programming model at present and a "wonderful collaborator". He said:

This is like the entry of electricity into every household, an "unprecedented" moment of change that will completely transform our development methods.

During the one - hour live broadcast, OpenAI spent most of the time demonstrating the programming ability of GPT-5, including presenting a series of benchmark test results.

Companies like Cursor, Vercel, and JetBrains also shared their evaluations of the early tests of GPT-5.

Michael Truell, the CEO of the "AI programming" tool Cursor, praised it as "the smartest coding model I've ever used":

The team found that GPT-5 not only performs well and is easy to guide but also shows a unique personality that other models don't have.

It can not only detect hard - to - notice deep - seated errors but also run long - term, multi - round background AI agents to complete complex tasks - tasks that often baffle other models.

Guillermo Rauch, the founder and CEO of Vercel, believes that "GPT-5 is the best front - end AI model":

Our initial impression when using it on v0.dev is that it is the best front - end AI model, achieving top - notch performance in both aesthetics and code quality, truly unique.

It performs excellently at the intersection of complex computer science and artistry, marking a leap from simple code completion in the past to full - stack applications across devices and screens today.

Kirill Skrygan, the CEO of the traditional IDE giant JetBrains, said that "GPT-5