Author：Wall Street CN
OpenAI is so leaky, it's like a sieve, regarding the latest Spud...(Potato)News about the model has come yet again.
This "potato",It is the much-anticipated GPT-6.
According to sources, this "potato" is fully cooked and will be released on April 14th.
Sources familiar with the matter say this is a model that's completely geared towards AGI (Asynchronous Genius).
With a 40% performance boost, it completely outperforms GPT-5.4 in code, reasoning, and agent tasks.
Native multimodal, a single architecture handles text, audio, images, and video.
It also features a massive 2MB context window.
GPT-6 will become a super engine, responsible for completely integrating ChatGPT, Codex, and the Atlas browser into a unified intelligent agent.
That's right, it's the desktop-level "super app" that OpenAI has been talking about for a long time.
What's most eye-catching is OpenAI's internal positioning of this model.
This is AGI's "last mile," and they're going to cut everything to gamble.
Is GPT-6 coming?
The person who brings you a lot of insider information is Strawberry Bro @iruletheworldmo.
This guy has some real talent; big names like Peter, the father of lobster, Gavin Baker, and Jim Fan are all his fans.
Strawberry Bro excitedly stated that OpenAI's internal workings have been leaking information like a sieve lately, and he has obtained a lot of juicy details from them.
First, the reason OpenAI cut all its side branches was to concentrate all its resources on GPT-6.
In a previous interview, Brockman stated that they were about 80% complete on their journey toward AGI.
From the perspective of OpenAI employees, GPT-6 represents the remaining 20%.
How so? Let the data speak for itself.
A native multimodal model, yet it still achieved a comprehensive leap in benchmark testing.
It is said to be 40% better than GPT-5.4 in coding, reasoning, and agent tasks.
The context window also reached an astonishing 2 million tokens, twice that of GPT-5.4 and Opus 4.6.
In terms of pricing, it also continues OpenAI's "fine tradition" of $2.50 per million tokens input and $12 per million tokens output, which is basically not much more expensive than GPT-5.4.
If we compare it to Claude, then it's like having Mythos.(myth)It boasts a level of intelligence, yet charges only Sonnet-level pricing.
It is said that GPT-6's pre-training was completed on March 17, and post-training and security work have also been completed, so it can be launched at any time.
The planned release date for the coarse cloth was April 14th.
As word got out, more internal details about OpenAI and GPT-6 were revealed.
Since December 2025, OpenAI has been in a state of "programming red alert".
Brockman recently admitted on his podcast that OpenAI was so focused on topping the leaderboards that it was outmaneuvered by Anthropic in the programming field, losing a large number of users.
The explosive popularity of AI-based programming products such as Claude Code, Cowork, and OpenClaw made OpenAI realize that "it is indeed possible to achieve AGI using only text".
This forced Ultraman to a dead end, leaving him no choice but to grit his teeth and cut almost all non-core product lines.
The most important project to be cancelled was Sora, which started strong but ended abruptly, indirectly affecting OpenAI and Disney.(It's all over the news)The billion-dollar contract is completely off the table.
However, that's not all.
The latest news is that Ultraman is no longer performing; he's focusing all his attention on data centers, and security issues can wait!
Currently, the OpenAI security team has been assigned to the CRO.(Chief Risk Officer)under.
Meanwhile, the department responsible for products at OpenAI was renamed AGI Deployment.(AGI Deployment Department)This clearly demonstrates their ambition.
After a series of dramatic actions, Ultraman finally managed to produce a...(perhaps)This is a powerful weapon sufficient to counter Anthropic's might, the GPT-6.
However, some commenters pointed out that Strawberry Bro's revelation may not be entirely accurate.
However, some people have come forward to support the claim, saying that although the specific details are questionable, the general direction should be correct.
True GPT Image 2
There's still no definite answer as to when GPT-6 will arrive, but GPT-Image 2 is definitely coming.
After all, it had already made a brief appearance at Arena yesterday, causing quite a stir as soon as it appeared.
Why? Take a look at these pictures and you'll understand—
Friends, I didn't post the wrong picture, nor was I slacking off while playing Minecraft.
This thing... was actually generated by netizens using GPT.
It can basically replicate any game 1:1, without any of the blurriness of AI, making it impossible to distinguish between the real and the fake.
And this Windows desktop, I was stunned for a long time when I saw it, wondering why this person would post a screenshot.
Then it dawned on me, oh, this was generated by someone using GPT-Image 2.
If the prompts were clearer, GPT-Image 2 could directly take over the YouTube homepage.
Its global awareness has also been greatly improved, completely aligning it with Nano Banana Pro.
The aesthetics are also quite good; it's not the typical bright blue sci-fi AI color tone found in raw image models.
The resulting diagram of human anatomy looks just like the illustrations in a textbook.
The realism has also been greatly enhanced.
Finally, the ugly yellow filter is gone, and the colors look much more normal.
We're really looking forward to it. If its performance is truly this stable, it will undoubtedly become the most practical raw image model to date.
Unfortunately, this model was removed from Arena yesterday and cannot be tested for the time being.
Computing power is the real manipulator.
Ultimately, all the models in AI competitions today point to one thing: computing power.
Moreover, its importance has become fully apparent.
Anthropic's decision to discontinue licensing for OpenClaw subscribers is not only paving the way for its own KARIOS, but also likely a move born of necessity.
I really can't hold on any longer.
The recent rapid depletion of tokens is probably due to this reason; OpenClaw has to take the blame, which has made me anxious about tokens.
Now, the cancellation of Sora and the termination of the Disney contract are both helpless moves made by OpenAI to make way for the computing power needs of new models.
Last year, when people talked about data centers, it seemed like a distant issue, much like environmental protection.
Now, the shockwaves from infrastructure improvements have spread along the industry chain to the application end.
This match is getting more and more exciting.
Given the scarcity of computing power, even a CEO like Altman, who is good at fundraising, cannot leave a way out for OpenAI.
It's a battle of who dares to take a gamble and bet on the one and only path to the future.
Reference links: [1] https://x.com/iruletheworldmo/status/2040508733944668408 [2] https://x.com/benitoz/status/2040524473233985565 [3] https://x.com/marmaduke091/status/2040338311873515597?s=20
This article is sourced from:
