Replying to @TokenGremlin
Why are you exaggerating so much? The truth is, the desktop app has been steadily getting better.
51
Honestly, GPT-6.2 Sol and Astra felt like trash to use on launch day.
I gave them another shot a few days later, and they felt a bit better. If you bounced off them on day one, I think theyβre worth another try.
18
Theoμ κΈμ μ½μ΄λ³΄μ
¨μΌλ©΄ ν©λλ€.
μ λ μμ§λμ΄κ° μλλλ€. GPTλ₯Ό μ λ§ μ’μνκ³ , μΌμκ³Ό μ
무μμ μ μ¨μ€λ μΌλ° μ¬μ©μμ
λλ€. κ·Έλμ μ΄ κΈμ κΈ°μ μ μΈ λ΄μ©μ μ λΆ μ΄ν΄νμ§λ λͺ»ν©λλ€.
νμ§λ§ κ°μ₯ ν¬κ² 곡κ°ν λΆλΆμ λΆλͺ
ν©λλ€. Opus 5.5λ₯Ό λ μ νΈνκ² λλ μ΄μ κ° λ¨μν μ½λ© μ±λ₯μ΄ μλλΌ, μ¬μ©μμ μλλ₯Ό μ΄ν΄νκ³ μΌμ λκΉμ§ λ§λ¬΄λ¦¬νλ λ₯λ ₯μ΄λΌλ μ μ
λλ€.
μ€λ μ μνΈλ¨Όλ λͺ¨λΈμ΄ λλν κ²κ³Ό λ³κ°λ‘, ν¨κ» μΌν λμ μ¬μ©κ°μ λ κ°μ ν΄μΌ νλ€λ μ·¨μ§μ κΈμ μ¬λ Έμ΅λλ€.
μ κ°μ μΌλ° μ¬μ©μμκ²λ μμ΄μ νΈκ° μΌμ μνλ κ²λ§νΌ, μ λ§μ μ΄ν΄νκ³ μ κ° μνλ λ°©ν₯μΌλ‘ μμ§μ΄λ κ²μ΄ μ€μν©λλ€. 무μμ νλμ§ μ κ° μ΄ν΄ν μ μκ² μ€λͺ
νκ³ , μ κ° νλ¨ν μ μλ λ΄λΆ λ¬Έμ λ₯Ό λ€μ μ μκ² λ λκΈ°μ§ μλ κ²κΉμ§μ.
μμ¦ GPTλ₯Ό μ°λ©° λλΌλ κ°μ₯ ν° μμ¬μμ μ±λ₯μ΄ μλλλ€. μ±λ₯μ΄ λ€μ²μ§λ€κ³ μκ°νμ§λ μμ΅λλ€. μν΅μ λ¬Έμ μ
λλ€.
λλνλ€λ κ°ν보λ€, βλ΄ μλλ₯Ό μμλ£κ³ ν¨κ» μΌμ λλλ€βλ κ²½νμ λ μμ£Ό νκ³ μΆμ΅λλ€.
30
Replying to @feedthejim @Angaisb_
Sorry to bring up a competitor, but do you know what makes Claude Code so good?
Whenever it finishes a task, it often proactively suggests what to do next and asks something like, "Here's what I think we should do next. Would you like me to proceed?"
As a user, all I have to do is read the suggestion and say, "Yes, go ahead."
I feel like Codex is still lacking in this area. When using the desktop app, I find myself having to give way too many instructions just to keep things moving. The workflow feels fragmented, with constant interruptions requiring human input.
Of course, Codex does suggest prompts when I press Tab, but those suggestions often seem to miss the actual context of what I'm trying to accomplish.
Claude Code, on the other hand, seems much better at understanding my overall intent, figuring out what needs to be done, and suggesting the right next steps based on that context.
I genuinely don't understand why there's such a noticeable difference. Is it a difference in the underlying models, the agent harness, or something else?
I'd really appreciate it if someone could explain what's causing this difference.
1
73
I shared my disappointment with OpenAI earlier, but there's still something I want to say.
This moment reminds me of when the Codex app came out about nine months ago. Back then, Claude felt ahead to me, while GPT seemed to be falling behind.
But with the arrival of Codex and improvements to the models, I watched OpenAI catch up with Claude quickly. Looking back, could today's competition and challenges also become the starting point for another leap forward?
They say a crisis can be an opportunity. OpenAI certainly has the capability. I'm hoping they'll show us something soon that turns the tables again.
53
Replying to @feedthejim @Angaisb_
I'm on Windows, and switching back and forth between local Codex threads and my dot takes way too long. I wish the transition were much smoother.
2
371
For the first time since I started using GPT, Tibo's reset and the banked reset haven't made me feel excited. I've been subscribing to the $200/month ChatGPT Pro plan for almost eight months.
I know OpenAI is launching new features like Wallet and gradually improving existing ones. But I don't think those changes have reached the point where they dramatically improve people's satisfaction. I understand how difficult this work is and that it takes time. It's unfortunate that consumers don't seem to have the patience to wait that long. That includes me.
Watching the competition makes the disappointment harder to ignore. For me, it's been a while since an OpenAI announcement made a really big impact. The 6.2 Sol I tried during testing was clearly better than 6.1 Sol, but compared with Claude 5.5, it didn't feel like a huge leap.
Does OpenAI feel the same way, and is the model taking longer to launch because they're still improving it? Or are there other challenges we don't know about?
I was hoping the 28-day reset journey would end with a major announcement. That was my expectation, not an official promise. But the longer the wait goes on, the more I wonder what's happening.
Having cared about GPT for so long makes this stretch feel even more frustrating.
1
203
Replying to @b_66352558 @thsottiaux
Exactly. Codex might be a fantastic tool for highly skilled developers who understand every aspect of their projects and know exactly what they're doing.
But that's not how most ordinary users work, right?
I feel like Codex is primarily designed with those experienced developers in mind. It might be an amazing tool for them, but naturally, for everyday users who don't have that level of technical knowledge or experience, it's going to feel like something is missing.
25
Replying to @thsottiaux
First of all, Codex feels a bit heavy. I know it has improved compared to earlier versions, but I still think there's room for improvement. The more I use it, the slower and heavier the app seems to get.
And there's one more thing, which I think is the most important issue.
I don't know if you've tried Claude Code, one of your competitors, but as an ordinary user, I can't really tell whether the difference comes from the models themselves or something else.
What I can tell you, though, is that the biggest difference I notice when actually getting work done is how smoothly the AI can handle a task from start to finish.
GPT doesn't feel quite like what I'd expect from an AI.
It can follow instructions accurately and do exactly what I ask. But I think a truly capable AI should be able to understand what I'm trying to achieve, even when I don't explain every detail perfectly. It should be able to figure out my intentions, come up with a plan, and take the initiative to get things done without needing me to spell everything out.
And I feel like GPT still struggles with that.
To me, that's the biggest difference between GPT and Claude Code.
1
1
61
Tibo is almost at 1M, and he just teased a possible banked reset on Thursday.
My guess: a reset could land in the next day or two. Stay tuned, friends π
teapaw.org/thsottiaux/status/2120β¦
mildly interesting but @thsottiaux is just 800 followers away from a million followers π€―
1
1
65
Ferroflora
A 16-second procedural sculpture made in Codex (6.2 Sol, creator-reported). The gold sphere represents a magnet. As it moves, the spiked metal folds into petals, then closes again.
2
2
100
Replying to @pashmerepat
I've been using this feature for a while too. There are still some ways it's less convenient than working directly on my local computer, but it's a powerful feature.
132
Last month, I shared a diagram imagining dot coordinating different tasks. Connecting multiple local computers was below the dashed line back then. Now it is a real feature.
Being able to choose the computer for each task is a step in the direction I had hoped for. Iβve updated the diagram to reflect that.
What Iβm curious about now is whether it can accurately track progress across tasks, check the results, and keep the work moving through to completion.
π€ Made with AI

ALT Personal concept infographic with a blue fuzzy dot in a black beret connected to cloud tasks and three local computers. The current capability is choosing a computer for each task. A separate dashed section labels collaboration with other dots as a personal vision, not an available feature. The workflow is Delegate, Review evidence, Continue or ask. Footer: Connection does not equal reliable coordination. A personal concept, not an official roadmap.
You can now connect multiple local computers to the same dot and choose which one it uses for a task.
It can build and test your app on a dedicated machine while you keep coding on your laptop.
2
57





