Token Gremlin @TokenGremlin

While merging Codex and ChatGPT into a completely new desktop app, OpenAI also renamed the old ChatGPT app to “ChatGPT Classic.”

One interesting detail in today’s GPT-5.6 launch: ChatGPT Plus users are also getting access to Sol. Not Sol Pro, which remains for higher tiers, but GPT-5.6 Sol through medium and higher effort settings in ChatGPT. That matters because Sol is the flagship model in the new GPT-5.6 family, and OpenAI is not keeping it entirely behind the most expensive plan.

OpenAI launched GPT-5.6 and ChatGPT Work, making the AI superapp thesis more concrete. The main change is not limited to model performance. ChatGPT is being reorganized into separate but connected surfaces for conversation, work execution and software development. Chat → reasoning, search, conversation Work → documents, spreadsheets, slides, reports, sites, web apps Codex → code, repositories, reviews, terminals, engineering workflows ChatGPT Work is powered by GPT-5.6 and can act across apps and files, remain active on complex projects for hours, and turn broad goals into finished deliverables. The Codex app is also merging into the new ChatGPT desktop app. The result is a more unified product strategy: ChatGPT is being positioned as a workspace for tasks, files and engineering workflows, rather than only as a conversational assistant.

A few weeks ago, I said that GPT-5.6 would arrive in a launch window close to the SuperApp. I did not imagine it would be ON THE SAME DAY.

OpenAI @OpenAI

2 hours ago

Today. 10am PT.

784 752 8K 1.0M 644

We are finally getting the SuperApp, or at least a preview of it.

This is the first AI image system I have seen generate an image in real time using geometric shapes, almost as if it were painting by hand. Watching the composition emerge step by step instead of appearing all at once is genuinely fascinating. It gives image generation a very different feel, much closer to a visible creative process than to a black-box instant output.

Reve @reve

2 hours ago

Reve 2.1 is here. The world’s best 4K image model just got better. Greater prompt understanding, world knowledge, and stronger foreign-text rendering.

42 77 512 64K 178

Another strong Grok 4.5 test. In this case, the model was compared against GPT-5.6 Sol, using an older checkpoint, and Fable 5, and still managed to stand out in this specific task. It produced the result in around 8 minutes with roughly 900 lines of code. The key point is execution reliability: the scene worked on the first run, with usable textures, controlled lighting, and no major runtime errors. There are still small spatial and orientation issues, such as misplaced objects, but the overall result shows that Grok 4.5 is becoming a highly competitive option for fast interactive prototyping. Cc: @mirochill

GPT-5.6 had already been ready for about three months and was being tested by several companies and reliable people. GPT-6 is literally ready as well and is being tested right now, and it is supposedly far superior to Fable in everything. Well, I am looking forward to the news.

We are reaching a point where most AIs are becoming genuinely good to work with. The differences between them are increasingly about nuance, and each user will probably end up preferring one model because of its particular strengths. GPT-5.5, Opus 4.8, Grok 4.5, Muse Spark 1.1... they are all excellent now, depending on the domain.

x.com/i/status/20751…

Chong-U @chongdashu

7 hours ago

x.com/i/article/2075…

2 3 40 31K 51

This 2.5D Animal Crossing-style game was built using only Grok AI tools. Cursor Grok 4.5 handled the code, Grok Imagine generated the sprites, and Grok Video 1.5 was used for the animations. The most impressive part is the iteration speed. For game development, where testing, adjusting, and rebuilding ideas quickly matters so much, Grok 4.5 is proving to be a very serious tool. Cc: @chongdashu

This little game was created with Grok 4.5, and honestly, a lot of people underestimated this model. It shows fast iteration, Three.js rendering, PhysX physics, runtime IK, vehicles, weapon stowing, and smooth interactive movement. The speed alone already makes a huge difference. Grok 4.5 is looking seriously impressive. Cc: @AshConnell

I admit that seeing OpenAI stop squeezing every last drop out of GPT-5.x and finally start something new with GPT-6 is getting me excited. They did an absolutely sensational job with GPT-5.6. I loved it, but everything has a limit.

Some highly reliable sources told me that GPT-5.6 Sol has around 4 trillion parameters, and that the GPT-5.x series ends with it. GPT-6 is the next model OpenAI is currently working on, with a much larger parameter base, which will supposedly surpass even Mythos / Fable-class models by a wide margin. Well, I am excited to witness all of this. I hope it actually happens. And the predictions point to sometime later this year.

@elonmusk @SpaceXAI Fable-class model when?

Some GPT-5.6 Sol testers say it is better than Fable, others say it is worse, and others say it is about the same... Honestly, it is starting to look like some testers may have had skill issues, if anything. We will find out in practice, which is much better. Especially with the more advanced modes: Max and Ultra.

@RedStaR_1337 @thegenioo Exactly. And in many use cases, this improves the model’s performance dramatically.

To be honest, GPT-5.6 does not even need to be better than Fable 5. If it is just as good, that is already great.

@thegenioo Well, Fable is also Anthropic’s most expensive model. For it to be a fair comparison, right?

@thegenioo He said he tested GPT-5.6 Sol, but not GPT-5.6 Sol Ultra. There is still hope.

@OpenAI