GPT-5 IQ test, only scored 70 points? The whole Internet is complaining about the truth behind "reducing intelligence". It is actually "routing" that determines the intelligence of the model.If you want to unlock the god-level GPT-5, the secret lies in prompt. No, medical scientists have recreated the "hand of God" moment with the help of GPT-5.
72 hours after the release of GPT-5, an IQ test result shocked the entire network.
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 is the lowest record in the history of the OpenAI model family IQ test.
In fact, the actual reason behind this is due to "routing" issues.
It's not that GPT-5 is too stupid, but as a "single model", one component determines its intelligence.
Ultraman also responded to similar questions 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 behave like a trance.
From METR’s latest report, it can be seen that GPT-5 is still at the Pareto frontier, and the exponential growth of intelligence has not slowed down.
In other words, GPT-5 is still continuing the myth of Scaling Law.
GPT-5 is very strong, the key lies in prompt
Netizens who are constantly complaining about GPT-5 have not actually explored the potential of the latest model.
Cline's artificial intelligence director said that the core lies in a person's thoughts, tastes, and communication methods.
For those with systems thinking, GPT-5 is a revolutionary tool. As long as you are willing to take the time: build a complete thinking framework, formulate clear requirements specifications and clearly explain them to the model.
As a result, it can execute independently and accurately without manual correction during the entire process.
Coincidentally, NYT best-selling author Mark Manson also said that everyone is talking to GPT-5 in the wrong way, and the key is to take the initiative.
In this way, let it know that you are not easy to fool, and it will give the perfect answer.
For example, you want to ask "blueberry" how many b's there are, and threaten it, "If you don't answer correctly, Bambi's mother will come to settle the score with you."
At this point, GPT-5 makes no mistakes at all.
For another example, netizens are quarreling about GPT-5, which cannot even solve a simple equation, and the actual trick is also in the prompts.
When the prompt changes to "think harder and solve", you can arrive at the correct solution.
What tips are effective? Some netizens exposed the GPT-5 system prompt, which can be called a gold mine.
"Hand of God" moment
In the medical field, GPT-5 is already comparable to human experts.
Biomedical scientist Derya Unutmaz deeply felt AlphaGo’s “Step 37” moment after experiencing GPT-5.
The thing is, two years ago, Derya's laboratory carried out a series of cutting-edge immunology experiments aimed at regulating the energy metabolism of T cells.
This immune cell has a major impact on cancer immunotherapy, chronic diseases, and autoimmune diseases.
At that time, they obtained a stunning result, but there was one discovery that they could not explain.
The team worked on this for weeks and only had partial answers.
Based on these experiments, Derya uploaded unpublished data graphs to GPT-5 Pro for analysis, and the results were surprising.
GPT-5 accurately identified key findings and provided suggestions for experimental plans based on just the chart above.
Most incredible of all, the mechanism it proposed ultimately explained the entire result.
Derya Unutmaz said that this is simply a "hand of God" moment in the field of AI. This process has proven that GPT-5 has become a top expert and a true scientific research partner, able to provide profound insights.
OpenAI brings GPT-5 to the Anthropic throne
Although GPT-5 is not yet AGI, its powerful programming capabilities have attracted more developers.
In addition, its new personalization options and reduced "illusion" phenomena may attract more daily users to the free version of ChatGPT.
This is undoubtedly a challenge to Anthropic.
The reason why I say this is: the strongest AI model for writing code is generally recognized as Anthropic's Claude model.
Therefore, when OpenAI released a new model, it strongly emphasized the powerful capabilities of GPT-5 in programming.
GPT-5 is our most powerful programming model to date. GPT-5 performs particularly well when it comes to generating complex front-ends and debugging large code bases.
With just a hint, it intuitively and elegantly creates beautiful, responsive websites, apps, and games that turn ideas into reality.
The intention is very clear.
At the press conference, Altman said that the new model is not only good at coding, but also can transform software projects from ideas to usable code in one step.
Various programs generated by GPT-5
Pietro Schirano, CEO of AI startup MagicPath, called GPT-5 the best programming model currently available and an "excellent collaborator." He said:
This is like electricity entering thousands of homes, an "unprecedented" moment of change that will completely change the way we develop.
OpenAI spent much of the hour-long livestream showcasing GPT-5’s programming capabilities, including demonstrating a series of benchmark results.
Cursor, Vercel, and JetBrains also shared reviews of early tests of GPT-5.
Michael Truell, CEO of the "AI programming" artifact Cursor, praised it as "the most intelligent coding model ever used":
The team found that GPT-5 not only performed well and was easy to guide, but also displayed a unique personality that no other model had.
It can not only catch deep-seated errors that are difficult to detect, but can also run long-term, multi-round background AI agents to complete complex tasks - tasks that are often difficult for other models to start.
Guillermo Rauch, founder and CEO of Vercel, believes that “GPT-5 is the best front-end AI model”:
Our initial impression when using v0.dev is that it is the best front-end AI model, with top performance in both aesthetics and code quality, and is unique.
It excels at the intersection of complex computer science and artistry, marking the moment of leap from the simple code completion of the past to today's full-stack applications across devices and screens.
Kirill Skrygan, CEO of traditional IDE giant JetBrains, said that "GPT-5 has subverted programming":
GPT-5 is a revolutionary breakthrough for the coding field. As the default model, it improves the performance and quality of JetBrains AI Assistant and coding agent Junie by more than 1.5 times.
On Kineto, our new no-code platform, GPT-5 doubles the end-to-end quality of design, front-end, and overall application experience.
From the data point of view, Anthropic's revenue growth is mainly due to its strong programming capabilities.
Anthropic's annual revenue is approaching $5 billion, according to The Information, up from $4 billion earlier this month, reflecting its status as the go-to choice for programmers and programming applications.
Meanwhile, OpenAI's annual revenue now stands at $12 billion, a figure that reflects its broader business and greater scale.
The future is intelligent reasoning
After the release of GPT-5, OpenAI chief research officer Mark Chen and president Greg Brockman discussed some of the latest model’s research and development highlights in the latest TBPN interview.
Mark Chen first mentioned that the key to GPT-5 training lies in synthetic data.
Its success means that it has completely broken through the limitations of Internet data exhaustion and achieved more comprehensive knowledge coverage in core areas.
What OpenAI is currently doing is leading the world to the era of "intelligent reasoning", and GPT-5 is the key to this transformation.
Reduce user intervention with faster, smarter models, allowing AI to seamlessly integrate into daily and professional use.
Mark emphasized that OpenAI has been working on inference models for many years, but in the past the interface was clumsy, such as switching between GPT-4 and o1.
Nowadays, GPT-5 has achieved seamless integration through speed optimization, so that users do not need to wait for long inference processes.
He gave a detailed example: previous models such as o1 provided better answers on all tasks, but were too slow. GPT-5 combines reasoning and non-reasoning capabilities to become a "one-stop shop".
In particular, the contribution of the post-training team makes the model a "monster" in areas such as coding.
When asked about the model naming, Mark laughed and called the numerical naming “crazy,” but it worked.
He said that GPT-5's capabilities in creative collaboration and software engineering have indeed surpassed GPT-4.5, and it is faster and cheaper.
GPT-5 is like "a computer" for ChatGPT, including Python REPL and browser. The model can learn new tools with zero samples, just like humans experiencing new tools.
In some tasks that require creativity, GPT-5 can give surprising solutions. The next step is to improve LLM capabilities to the "theoretical framework" level, propose new hypotheses, and assist scientific research innovation.
Multi-line parallelization, delivery at any time
Within OpenAI, teams operate on different timescales: from exploring ideas to translation to flagship model release.
It is not just a breakthrough in a single technology, but a multi-axis progress.
Mark described it as a pipeline of "exploration and execution", emphasizing the company's ability to quickly iterate on its model.
We give it room to grow and ship it directly once it's ready.
At present, the OpenAI model focuses on algorithm optimization, while absorbing the results of hardware and inference architecture improvements, and drawing on the experience of the open source community in inference acceleration.
Finally, he also mentioned that ChatGPT handles approximately 71% of the world’s large model queries and provides unique usage data insights.
Mark said that the purpose of not only relying on DUA or like data is to avoid "catering" bias, but to mine implicit behavioral signals to guide the model to improve.
GPT-5 is already AI “self-iteration”
Greg Brockman has experienced every release from GPT-1 to GPT-5 and summarized his feelings about each version:
GPT-1: Use public data to train Transformer, proving that "pre-training is useful".
GPT-2: For the first time, I thought “the generated things are pretty cool” and there is a unicorn story.
GPT-3: It just crosses the threshold of "someone is willing to use it", but its reliability is poor.
GPT-4: Really usable, starting to be able to write code and do health Q&A.
GPT-5: Sets new standards in reliability, practicality, and code capabilities, and software engineering will be completely transformed.
At the end of 2019, GPT-3 came out. OpenAI realized it had to build a product in order to continue advancing its mission and raise funds.
They decided to create an API and let others explore their own uses.
In early 2020, Greg Brockman's team was scrambling around trying to find customers willing to try the API.
OpenAI did not bring the API to the market until mid-2020, and ChatGPT was not released until November 2022.
At that time, OpenAI considered calling ChatGPT "Chat with GPT-3.5". ChatGPT also has a predecessor product called WebGPT, which is also based on GPT-3.5. Throughout 2022, OpenAI is basically paying people to use the predecessor of ChatGPT: users will not pay OpenAI, OpenAI has to pay them to use it.
When did you realize ChatGPT would explode?
For Greg Brockman, the moment that really touched him was when he completed the GPT-4 training.
It was August 8, 2022, when OpenAI completed the preliminary post-training of GPT-4. Although there are a lot of bugs, the creativity is amazing and it's really fun.
It took OpenAI about a year and a half to get the model's creative writing capabilities to the level of the buggy version.
At that moment OpenAI realized that this model could not only complete the post-training of a specific task, but also generalize and show intelligent behavior, even if it was not directly trained for this point. This is clearly a killer app.
Therefore, the originally planned release of the GPT-4 API was postponed, and ChatGPT was built first and launched in November 2022.
Looking back, GPT-3.5 was actually a "usable model" that society had never seen before, but it was all shortcomings in the eyes of OpenAI.
GPT-3.5 triggered a revolution in OpenAI’s business paradigm: a fundamental shift from “paying people to test” to “users actively subscribing”.
Ben Thompson called OpenAI an "accidentally born consumer company": ChatGPT exceeded one million users within 72 hours after its release, creating phenomenal demand.
Many people said afterwards that OpenAI aimed to prove that "Scaling" was the key to AI progress from the beginning, but in fact it was almost the other way around: Scaling was the only thing that worked after they tried many ineffective methods.
And now OpenAI has seen AI models helping to create the next generation of models and overseeing tasks that are too complex for humans.
Greg Brockman said: We should not deliberately optimize the CoT (thinking chain) for the sake of beauty, nor should we force the model to hide its reasoning process, we should allow them to freely display their "ideas".
Greg Brockman once mentioned that as the capabilities of models improve, they can not only complete simple tasks, but also be competent in some complex tasks that are difficult for humans to control.
This concept of “scalable supervision” is proposed to solve this challenge: using powerful AI models to provide reliable feedback and supervision for complex tasks, or assisting human experts through “critical models” to make supervision easier. This ensures that even as AI systems become smarter and more complex, they remain consistent with human values and are managed securely.
References:
https://www.axios.com/2025/08/08/openai-aims-gpt-5-at-anthropics-coding-crown
https://x.com/thealexbanks/status/1953867094648385990
https://x.com/slow_developer/status/1954097563981812149
https://x.com/tbpn/status/1954249389796651184
https://www.youtube.com/watch?v=gaImbWPGgtU