GPT-5 benchmark test leaked, rumored to be released two days later?

Early in the morning, news about GPT-5 has arrived again.

These leaked GPT-5 benchmark tests are likely to be real.

There’s even a bombshell news: GPT-5 will be released on July 31.

As a result, all GPT-5 models have officially exited the WebDev arena.

However, there are other claims from Menlo Ventures investor Deedy, as well as foreign media like The Verge and The Information, suggesting that GPT-5 will debut in August.

Although GPT-5 hasn't arrived yet, practical tests about it are already all over the internet.

Just now, another user released a practical test of GPT-5 replicating the Minecraft game. To be precise, it is the GPT-5-pro with the internal code name zenith.

This user commented: "Impressive, it's simply magic! OpenAI has indeed created something incredible."

In this video, GPT-5 completed the game tasks fluidly in one go, and the performance is truly stunning.

With the expectations raised so high, GPT-5's official release is bound to be explosive; otherwise, it would be hard to know how to wrap things up

There is another heavy revelation from the well-known informant Jimmy Apple.

According to him, many internal evaluators have rated GPT-5 as being stronger than Grok 4 Heavy.

GPT-5 is coming, everyone holds their breath in anticipation

Now, the arrival of GPT-5 is getting closer.

Some have even discovered that when they select o3 in the app, they unexpectedly tested a certain version of GPT-5.

More and more people are accidentally testing GPT-5.

The news about its launch this week has also been confirmed by more and more people.

However, The Verge has a slightly different take; according to their intelligence, GPT-5 will be released in early August, including mini and nano versions.

Previously, developers had discovered that GPT-5 was internally named "Inference Alpha Version."

At the same time, a model codenamed "o3-alpha" was taken offline just 12 hours after its launch, with many acknowledging that this was an early shell of GPT-5.

According to OpenAI's usual practice, the interval from testing to release is at least 4 days, so GPT-5 is indeed quite close.

Just yesterday, it was discovered that GPT-5 can be used on LMArena. The Zenith model was also found at the same time.

The following examples have already gone viral across the internet.

Generating a starship control panel from a distant future.

Create a streaming media website.

Perfectly present SVG animations in robot walking.

The best pineapple defense game in history.

Integration of o series and GPT series

There is no doubt that GPT-5 is now the most anticipated model in the world.

Many believe that GPT-5 will be a significant milestone that will attract millions of users to join the AI ecosystem.

Next, we will sort through the various clues regarding GPT-5 mentioned over the past period.

During a live broadcast about OpenAI's intelligence, GPT-5 was mentioned.

The key information at that time was: this amazing cutting-edge model will unify the two series of models for the first time, concentrating on the breakthroughs in reasoning from the o series and the breakthroughs in multimodality from the GPT series.

Because ChatGPT has various models, each with its unique functions and outstanding features, if GPT-5 is indeed a collection of the best parts of each individual model, it is clear that the user experience will be completely transformed.

For example, those who have used o3 know how crazy the leap from GPT-4o to o3 is.

This was confirmed as early as February this year by OpenAI CPO Kevin Weil.

A netizen asked: Will you create model routers, or will they be more unified in a systematic way? Weil stated that they will be more unified Additionally, there is a revelation from a suspected internal employee of OpenAI. He stated that researchers did attempt routing methods, but it resulted in many hallucinations.

Therefore, they are testing a model that can plan, reason, and utilize agents like an extension.

Next are some revelations from the foreign media The Information.

In summary, GPT-5 has extremely strong coding capabilities.

In the field of natural sciences, reasoning is deeper;

Automatically completes complex tasks in the browser;

Writing is smoother, and logic is more coherent;

More importantly: there is a tremendous improvement in coding!

According to one tester, GPT-5 is not only better at solving academic and programming competition problems but also performs even more impressively when handling actual programming tasks faced by real-world engineers.

For example, it can modify large codebases filled with legacy code without any fear.

It is this meticulous ability to handle complex scenarios that has kept OpenAI's models behind Anthropic in the past. After all, among the developer community, everyone agrees that Claude is the true king of programming.

One tester stated that GPT-5 even directly surpasses Anthropic's Claude Sonnet 4 in programming!

Another claim is that GPT-5 is not a unified model but a routing mechanism.

It will send your questions to either a GPT large model that excels at casual conversation or an o-series model that specializes in logic and reasoning, depending on the type of question.

Ultimately, what we see as GPT-5's performance is the combined effect of these two models.

There are even OpenAI executives privately predicting—

We are confident that we can achieve GPT-8 without changing the architecture.

This means that OpenAI does not intend to roll out a new architecture but rather aims to maximize existing technology step by step through smarter scheduling, stronger reasoning, and more post-training data.

What will GPT-5 bring to the world?

At the same time, Ultraman's recent statement in an interview about "GPT-5 making him feel useless" has raised expectations even higher.

Some people say that GPT-5 is likely one of the most dangerous things happening in the AI field right now.

For example, Ultraman mentioned in this interview that many people spend all day chatting with AI, even treating it as their boyfriend or girlfriend.

There are also some children who rely entirely on scrolling screens to obtain dopamine during their growth. These situations are very dangerous.

When the host asked: How to prevent AI from having the same negative impact as social media? Ultraman honestly admitted: I am very afraid of this, and I have no answer.

What is concerning is that just a few days ago, an investor from OpenAI admitted that he has experienced some abnormal conditions due to using ChatGPT all day.

In other words, even wealthy people can develop mental health issues from chatting with AI.

Ultraman even expressed great interest in providing free access to GPT-5 for everyone on Earth.

When these AI products and services are offered at 1/100 of the cost, it is clear that certain economies will rapidly transform and collapse.

However, regardless of the kind of turmoil it may bring to the world, the momentum for GPT-5 to go live is now unstoppable.

Author of this article: Xinzhiyuan, Source: Xinzhiyuan, Original title: "GPT-5 Benchmark Test Leaked, Expected to Launch in Two Days? Shocking Minecraft Recreation Causes Netizens to Call it Divine"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at their own risk