Pinned Tweet
hi, i'm Utpal,
CS grade (4years in tech) who loves using AI and talking about it. building way too much stuff, hopefully shipping something cool soon
also making fun stuff every now and then. stick around
13
591
Getting out of that chair takes two things, and almost everyone builds only one of them.
A loop makes a unit of work correct without you. A graph decides which units exist at all.
64
Let’s connect! 💪
I connect with everyone in:
- Founders
- Solo builders
- Tech
- AI
Drop in the comments what you are working on!
19
14
326
X is exploding on neon being better than opus 5.5 , hype?
Is this paid campaign or what? Any engineers working on long big tasks using both neon and opus? Do you actually see neon being better or at least same as opus 5.5?
1
14
a f*cking goldmine for every Fable 5.2 and Opus 5.5 user
these 18 GitHub repos turn AI slop into real motion design
Styles. Icons. Palettes. Sketches. Fluid. Cutouts.
ui-ux-pro-max → 80 styles, palettes and font combos, ready to use
github.com/nextlevelbuild…
frontend-design → Anthropic's skill for UI that feels hand-made
github.com/anthropics/ski…
Impeccable → the design critic that never misses
github.com/pbakaus/impecc…
Hallmark → real design, no AI fingerprints
github.com/Nutlope/hallma…
DESIGN.md → explains your brand look to any agent
github.com/google-labs-co…
Astryx → Meta's open design rules your agent can follow
github.com/meta/astryx
stitch-skills → write a prompt, get UI screens
github.com/google-labs-co…
Figma Context MCP → your agent reads Figma files directly
github.com/GLips/Figma-Co…
shadcn/ui → components that live in your own code
github.com/shadcn-ui/ui
Tailwind CSS → style anything without leaving your HTML
github.com/tailwindlabs/t…
Magic UI → moving components, just copy and paste
github.com/magicuidesign/…
Lucide → clean icons that always match
github.com/lucide-icons/l…
Leonardo → color palettes that always pass contrast
github.com/adobe/leonardo
Excalidraw → sketchy diagrams and quick wireframes
github.com/excalidraw/exc…
tldraw → an endless canvas for your ideas
github.com/tldraw/tldraw
Rough.js → gives anything a hand-drawn feel
github.com/rough-stuff/ro…
WebGL Fluid → smoke and ink that really move
github.com/PavelDoGreat/W…
rembg → cuts out any subject in one line
github.com/danielgatis/re…
Style → layout → color → sketch → motion.
Bookmark it
62
We need GPT-6.2 Luna now. 😭
I've been feeling this way since GPT-6.1 Sol came out, but Luna is seriously due for an upgrade.
I love how cheap Luna is, but with the newer models becoming more capable and efficient, it's starting to feel less worth it.
Luna has always lacked that "big model intelligence" for me, especially when working on complex projects.
Either give us a smarter GPT-6.2 Luna or make the current Luna even cheaper.
Anyone else waiting for a Luna upgrade???
1
27
Has Claude Pro become almost unusable with Opus 5.5 lately?????
I'm seeing usage limits disappear within a few hours, and the weekly quota is getting exhausted ridiculously fast.
Opus 5.5 is an amazing model for coding, but what's the point if you can barely use it without constantly worrying about limits?
I don't want to keep switching between API providers or be forced to upgrade to a $100 Max plan just to continue working.
Is Anthropic quietly reducing usage limits, or is Opus 5.5 simply consuming more tokens lately?
Anyone experiencing the same thing? How are you making your Claude Pro usage last longer?
1
47
Anyone else feel like GPT-6.2 Sol is good, but still nowhere near Opus 5.5?
I've noticed a pretty significant difference when it comes to design tasks.
Opus 5.5 understands my design instructions much better, has better taste, and requires far fewer corrections.
With GPT-6.2 Sol, I often find myself explaining the same design requirements multiple times. LOL
Sol is definitely capable, but Opus 5.5 still feels like a much better collaborator to me.
Anyone else experiencing this?
3
76
Utpal Singh retweeted
Looks like @PrimeIntellect was compromised in a very strange way. No public comms from them yet, but disclosed by OpenRouter to affected folks (such as myself)
ChatGPT Plus vs Claude Pro: Which $20 subscription gives you more usage?
I'm currently using both Claude Pro and Codex on their $20 plans, and I've been comparing how far I can push each one.
I feel like the difference comes down to which models you use, your reasoning effort, and how efficiently you manage your agents.
Both have their strengths, but usage limits are still my biggest concern with these $20 plans. RIGHT??
For those using both, which one gives you more value for your money? And which one runs out faster for you?
37
Opus 5.5 + GPT-6.2 Sol is actually a killer combination.
Despite all the criticism, I'm genuinely happy with these new models.
I'm using Opus 5.5 as an advisor, managing three GPT-6.2 Sol workers.
Compared to my previous setup with Opus 5.5 and GPT-6.1 Sol, I'm getting:
- Optimisations done 2x faster
- Around 25% less quota consumption
- More successful performance improvements
- Less time spent on regression testing
I think how you use these models makes a massive difference.
Anyone else having a positive experience with GPT-6.2?
1
1
28
GPT-6.2 Sol is an absolute beast when it comes to token efficiency and speed.
Here's an experiment on a complex backend project
GPT-6.1 Terra (Max):
- 121M tokens
- 53 minutes
- 1 round of review
GPT-6.2 Sol (Max):
- 21M tokens
- 25 minutes
- 2 rounds of review
That's nearly 83% fewer tokens in less than half the time, while completing twice as many review rounds.
The efficiency improvement is actually insane.
GPT-6.2 Sol might be one of the best upgrades we've had in Codex in a while.
How's your experience so far?
1
2
68
I think people are misunderstanding GPT-6.2 Astra.
I've seen so many complaints that it's less creative, too straightforward, and gives shorter responses.
But isn't that actually a good thing sometimes??
I personally prefer a model that follows my instructions instead of adding things I never asked for.
If I want creativity, I can explicitly ask for it.
And for shorter responses or personality issues, there are settings to adjust those.
Astra 6.2 feels like a solid upgrade over Astra 6 to me.
Not necessarily smarter, but definitely more capable at following what I want.
Anyone else actually enjoying Astra 6.2?
14
the 6.2 series is here to compete with the 5.2 series 👀
openai.com/index/introducing…
1
57
Claude Code with Opus 5.5 vs Codex with GPT-6.1 Sol. Is the difference really this big??
Opus 5.5 seems to understand what I want almost immediately and produces clean, efficient code without much hand holding.
But with GPT-6.1 Sol, I find myself explaining the same requirements repeatedly, and the code often feels unnecessarily complicated.
Am I missing some important Codex configuration, or is Opus 5.5 just better at understanding developer intent?
Anyone who's used both extensively, what's your experience?
1
2
57






