GPT-5 release node is advancing again! On LMArena, the "super-sized" GPT-5-pro, with internal code name zenith, has been quietly launched. Some lucky users have started impressive testing.
Rumors suggest that the release of GPT-5 has been brought forward, expected to debut by the end of this month.
Users have revealed that GPT-5 has 4 versions:
GPT-5-pro (zenith)
GPT-5-high (summit)
GPT-5-mini
GPT-5-nano (starfish)
On Saturday, users prematurely released some testing of OpenAI's open-source model with the code name "lobster".
Today, more stunning demos of multiple GPT-5 versions have been revealed.
On Reddit's hot list, the mysterious model summit created an online video website, attracting many user comments.
First User Test, Game Websites Covered
Currently, GPT-5-pro (zenith) has been quietly launched on LMarena.
A developer used the most powerful version zenith to generate a "Nebula" single-page website with impressive speed and quality.
A user who drew zenith on LMarena successfully added a battle feature to Tetris.
There are also some tests of the GPT-5-high (summit) version, with users believing it is much smarter than us.
It can generate a spaceship control panel from a distant future.
It can build a video website.
GPT-5-high (summit) can also precisely generate SVG images of robot walking.
It can even create the best pineapple defense game.
And a game about UFOs abducting cows.
Besides these exposed tests, some US companies received GPT-5 preview versions last week.
Now, more and more evidence suggests that GPT-5 is about to emerge.
With GPT-5's Emergence, Has Programming Reached Its Limit?
A week ago, Altman posted an intriguing post on X:
Woke up and tried our new model for a programming project, completed in 5 minutes. It's incredibly, incredibly powerful. I don't know how to feel...
Although not explicitly mentioning GPT-5, the timing coincides with OpenAI's typical flagship model summer release window.
Industry analysts point out that programming capability has always been an important benchmark for AI model performance.
As emphasized in Stanford's "2024 AI Index Report", existing models can only solve 15% of programming competition problems, with obvious shortcomings in complex programming tasks.
The emergence of GPT-5 may completely rewrite this situation.
According to industry rumors and partial internal leaks, GPT-5 is expected to bring a series of notable new features:
- Autonomous Task Handling: Context-aware model switching without user intervention
- Enhanced Multimodal: Upgraded image/video analysis with reasoning capabilities
- Enterprise-level Programming: Rapid prototype development with debugging functionality
- Memory Expansion: Deep cross-dialogue context recall
As GPT-5 is about to reshape programming, research, and problem-solving patterns, one fact becomes increasingly clear: AI not only accelerates tasks but is redefining the boundaries of human capabilities.
References
https://x.com/Prashant_1722/status/1949313531356094558
https://inews.zoombangla.com/gpt-5-release-altman-coding-reasoning/
This article is from the WeChat official account "New Intelligence", author: Peach, published by 36Kr with authorization.
