>>109255567
r8 my website
https://cy-x.net/

>The user asked for an explanation, I should take a pedagogical approach

File: 1764351910874418.gif (934 KB, 128x128)

934 KB GIF

What's your ROI from vibing?

>>109255584
the hell?

>>109255610
Can I see some proof?

>>109255633
I'm selling ~3 web platforms per month around 5k for each.

>>109255648
That's pretty good.
>>109255633
I only make money using AI at my job, my personal projects don't make any money yet.

>>109255633
My goal with all of this is to make my job easier. I don't look at it in terms of money. I want to get to the point where I barely need to do any work yet still get my salary.

File: 817iJHQQXnL._AC_.jpg (237 KB, 1707x2560)

237 KB JPG

>>109255633
Last month I made 1.30 Euros from Google AdMob
I spend 50 Euros on AI/month (poorfag subs to OpenAI, Anthropic and Google) which are enough since I only do this on weekends (I wageslave during the week)
I also spent around 500 Euros to set up my business (registration, liability insurance) and 99 Euros to set up an iOS Developer account

So far I am -798.70
My MRR is -48.70

File: file.png (82 KB, 790x774)

82 KB PNG

Sol Ultra worked since Friday morning on one of my biggest/hardest features. There's a lot of compilation I'd estimate actual clanker time to be half or less.
You might get the gist from the screenshot but
>project is ml inference engine, similar to TensorRT or AITemplate
>models compile to artifacts or modules
>think like model as 20mb self contained module vs several gb for entire pytorch venv
>artifacts/modules have an internal model runtime, and the Python session runtime interfaces with that
>things like tokenizers, image processors, loading/packing weights, etc. is in python
>feature moves everything out of python, rust runtime
>clanker implemented entire image processors library in rust, compatible with all the image processors in diffusers and transformers
>clanker implemented webdataset in rust
>used safetensors and tokenizers, already written in rust
>complete end to end pipeline
He optimized things as he went too, and fixed some things when preparing to push
Next is to make it more generic so it can easily fit with other models as I integrate them

>>109255737
Do you do this with only unit tests, or do you have some end to end data? E.g. do you test that the predictions are still the same within some tolerance?

File: big-pickle-fuckup.png (373 KB, 1563x2798)

373 KB PNG

>>109255391

For anyone who wanted it, here is the patch script to build a local version of Desktop Commander. I hope I'm not leaking anything private.

https://litter.catbox.moe/h0wyj1.sh

Also, my retarded ass might have almost lost the changes by messing around with the free model on OpenCode while trying to generate the patch from the modified files (but I also think maybe OpenCode shouldn't let the model run dangerous commands like git checkout without an approval prompt by default, especially when shipping by default one of those small models that love to revert everything and lose uncommitted changes). I forgot that's a thing with open weight models. And the esc key doesn't work half the time on tmux.

>>109255633
many thousands of dollars deep into the red
$0 revenue

>>109255762
For the models there's parity tests against the pytorch model. The image processors have checked in fixtures

Anyone use Cursor? Their Composer 2.5 fast seems to be pretty good.

>>109255633
more free time

>>109255811
i use claude design

I want to say that vibing is mentally exhausting sometimes. Even if I don't come up with the code I still need to come up with the plan. Just a basic description of it won't cut it because I'm exacting. Going from the idea in my head to the prompt can be such a chore. But it's still easier than what I was doing before.

testing the new Sites feature and sol UI
https://interface-atelier.zxc61.chatgpt.site

what are you all using to prompt images? mostly like website images/stock image replacement type stuff

>>109255633
+$150 (donations, - $140 (subs). There's IMMENSE hype for the premium version of my app, so I'm going to make at least a few grand with my launch.

File: Screenshot 2026-07-12 at (...).png (321 KB, 1720x1296)

321 KB PNG

>>109255787
Thanks anon. I think I found a better way to have "unlimited Sol" and easier. I am still grateful for your idea.

Instead of using Desktop Commander, use this code-focused local MCP https://oraios.github.io/serena/01-about/000_intro.html

Set up a tunnel in chatgpt, create an API key (for the tunnel) and wire it in with the local serena mcp. For questions you can ask me or better, ask Sol to wire it up like I did (also via opencode, lol).
Note: It's obviously rather dangerous but I don't mind as I run this shit in a lxc without sudo rights.

>>109255877
comfyui

>>109255836
>But it's still easier than what I was doing before.
I can vibe for 12-14 hours straight.
I was conked out after 6 hours of coding and basically had to stop for the day.

>>109255836
anytime you're being introduced to a new concept you have to take in it gets really exhausting really fast. it's a very strong streamlined learning process though.

>>109255937
I was up all night working on the current project and went to bed at 6am. I went hard but didn't feel great when I finally woke up at 2pm.

>>109255937
When I was programming normally I planned my day in such a way that I would do the easiest tasks last, and then after actual programming I would still spend some time answering emails and so on. That way I could work for quite a long time, but the most productive work was also around 6 hours I would say.
With AI I basically work on code the whole day, and if I have to answer mails or whatever, I can do that in parallel.
The downside is that I sometimes I stay awake for too long, because it's so easy I always think one more prompt won't hurt.

is there a good way to handle non-bias sentiment labeling with AI

>>109255959
yeah

>>109255957
But even now I do the hardest tasks first, basically planning new campaigns is usually done in the morning, in the evening it's bug fixes, or maybe a bit of frontend.

>>109255960
go on

File: 1770154993126128.png (234 KB, 686x386)

234 KB PNG

>Hi, I'm a big fan of your project, can you please add x feature?
>Yes, I'd love to do that. I'll get working on it
>Open notion, add it
>Open Codex (0 Resets left, 0 Weekly Allowance.)
PLEASE I WISH I HAD MORE TOKENS PLEASE SAM I'LL DO ANYTHING. I'LL SUCK YOUR DICK AND EAT YOUR SHIT!!!

I'm going to bite the bullet and pay 5$ to Cline to see what the fuss is about.
What should I build?
I actually have no fucking clue.

>install gentle-ai and engram
>now my prompts get stuck and never finish
Has anyone had the same problem?

>>109255945
I guess I'm speedrunning it all so it's abnormal. My autistic brain has been doing this at a much slower pace for 37 years and suddenly it's being bombarded with a flood of new concepts and possibilities.

>>109256027
what is gentle an engan?

I want to use video and audio generation to massively compress videos for archival in a lossy ways. suggestions?

>>109256005
use mcp with chatgpt

>>109256105
what platform?

>>109256136
i was thinking of locally archiving dev streams

File: gpt5.6-ui.png (131 KB, 1038x881)

131 KB PNG

GPT 5.6's web UI taste is absolutely horrible.
How many scam memecoin pages and fake software projects was it trained on for fucks sake?

>>109256027
I uninstalled both of them and it now works

Anyone using codex-lb? Is it stable again? Or any alternatives?

>>109256215
i like it

File: voice-licensing.png (70 KB, 632x793)

70 KB PNG

ChatGPT is SO FUCKING ANNOYING with the voice cloning stuff when working on TTS.

>>109255633
>my revenue: $0
>my costs: also $0
Good enough for me.

>>109256264
kek kwabGPT

>>109256215
the openclaw mobile UI was trashed hard because it was built with gpt

>>109256215
Better than before, but I still always use Claude for frontend.

>>109256329
hey how do you do that?

>>109256264
Are you using a 1366x768 monitor lmao
why does your text look so atrocious

>>109256449
nope. 1080p laptop.
I guess I resized it to take the screenshot idk sharpness is not a metric anyway

>>109256464
yes it is

File: spawnthinking.png (25 KB, 986x309)

25 KB PNG

Hmm, tokenizer mismatch, my favorite. (This happened after trying to regenerate a voice mode response in normal mode with ChatGPT)

File: file.png (331 KB, 540x576)

331 KB PNG

>>109256673

>>109256713
qrd on whatever the fuck you are doing

File: file.png (9 KB, 772x115)

9 KB PNG

I could have saved an entire week last week by using fast mode

>>109256779
I sometimes use it but it's expensive as fuck, 2.5x. With 5.6 I only use it for very few tasks.

File: bd.jpg (1.06 MB, 1788x2162)

1.06 MB JPG

Using up my Fable today and then I guess going to switch back to GPT, since not paying API prices. Stupid move Claude.

>>109256805
I spent 200 dollars in api today to wrap up, fuck claude

>>109256811
there just isn't any reason to remain to loyal to any platform. i'm generally not someone who tries to figure out the "AI Industry" but one thing that's for sure notable about it is that it's hard to really have a moat.

File: spawnthinking-full-schizo.png (886 KB, 798x8649)

886 KB PNG

>>109256725
I'm was using ChatGPT with MCP to work on some code. I switched on the new voice mode then tried to regenerate the answer in normal mode.
In case you care I kept probing for a few more turns so here is the conversation in full schizo detail. White is messages in normal mode, gray is voice mode.
The conversation was started in text mode and voice mode is set to High thinking ($100 plan).
The "one moment" thing reminds me of this.
https://www.youtube.com/watch?v=4s5yHUpumkY

I am assuming the chinks are distilling Fable at full throttle at the moment.
How many months until we get another mythos class model?

>>109256886
distilled models are never as good as the teacher model
we'll just get 5.5 maybe

>>109256886
sol works as good desu

how many hours until fable is gone

>>109256910
I think it's cali time

>>109256899
We can only use self hosted open models at work. So a 5.5 level open model would already be an insane productivity boost compared to GLM5.2 (which is good but not quite there yet)

>>109256910
Probably midnight Pacific Time

i'll probably have fable usage leftover when they take it don't really have any brilliant ideas right now

gemini 3.5 pro in 5 days

>>109256994
only 11% on polymarket
Google fucked up they can't reach Sol or Fable with their pro model

>>109256953
I just launch a bunch of fable subagents to find things to improve if I need to use my limits quickly.

I don’t like how in codex I have to do /plan for every prompt and it literally tries to create a plan. Whereas in opencode I can switch between plan and build mode and casually talk to the clanker.

>>109257049
I just do planning and execution in the "regular" mode whatever its called. I don't see why plan mode needs to exist at all.

>>109256264
just give it a sample of the voice you want bruh and just don't tell it it's scarlett johannson or morgan freeman or whatever, it can't actually hear it and it'll never ask

>>109257007
how did that happen? how did meta lose the open source race? what the fuck

>>109257049
What the fuck is /plan?

>>109257049
>I don’t like how in codex I have to do /plan for every prompt
Erm you don't have to? I used it like once

>>109257121
gated big design up front so you can catch mistakes
it’s generally better if you’re more on the “agentic engineering” end of the spectrum as opposed to “vibe-coding”

>Suddenly hear machine voice talking to me
>Get fucking scared
>Says something needs my attention in project
>What the fuck?
>5.6 Ultra decided to use Sam voice mid goal to ask me for something because I wasn't responding
>Never instructed it to do such things
Spooky

>>109257132
>how did meta lose the open source race?
no money in open source
I assume the only reason Chinese companies open source their models is some kind of CCP sponsored economic warfare against the US. I am watching from the sidelines here (EU with no AI company worth mentioning except maybe Mistral who lost even more traction).
But I can't think that those chinese companies actually financially benefit by open sourcing their models.

>>109257142
I can just ask it to create planning documents in the regular mode. I just don't see the point in another mode.

>>109257160
you euros are such assholes, so much money, so much people, where are the models???

>>109257134
ask your clanker

>>109257157
that would be creepy but I’ve thought of prompting my clanker to get my attention with say(1) when it’s all done

>>109257161
oh, plan mode explicitly forbids modifying files
might be useful if you want to control writes

>>109257165
There's no money here and arguably only a handful of countries in the EU can even be considered first world (Germany, France, Nordics, UK before they left). Have you ever been to Romania? New "candidates" to join the EU have a GDP per capita comparable to Botswana or Jamaica. Where are all the Jamaican LLMs?
Modern EU is hyperconservative (anti-technology) and hypersocialist in the worst possible combination. Also classist. There is very little upward mobility if you don't belong to the right social class from birth.

>>109257209
how can you be hypersocialist and classist?

>>109257165
What models can I train on my 5070TI?

File: 1401108178947.jpg (38 KB, 362x346)

38 KB JPG

Guys what agent orchestrator are you using? I want a GUI where I can start isolated async jobs and get a MR some time later but I am not familiar with existing work. Would openhands offer this, or some GUI over opencode, or..?

=>>109257249
sol ultra

>>109257249
just vibecode your own

>>109257235
you cant imagine the ideology which invented "class warfare" being classist or what is the question?
on the first day under socialism, everyone gets free everything
on the second day under socialism, being unemployed is made illegal and we kill the kulaks for leeching, so that we can keep everything free

File: 1758754538740968.png (317 KB, 960x540)

317 KB PNG

This mother fucker Sol Ultra has been running for 15 hours. The code/architecture is so clean even a retard could understand it now. But my usage... it's depleting fast. I haven't been able to use my code base in over two days.

File: 784ffeedf9fe1bd32cfd8fa13(...).jpg (190 KB, 1200x857)

190 KB JPG

Sometimes I start a long running task that will take 1+ hours and think I should go for a walk or do some chores around the house but I can't help but stare at the screen in amazement and wonder looking at the lines fly by, hypnotized by what the magic machine is doing.

>>109255567
GLM-5.2 makes a better OP than you.

why is the thread so obsessed with luddites? why even think about them at all? it's really coming off as a cope

>>109257336
it's because 'dites like you flock here to get berated and anally raped, it's a humiliation fetish your kind has ingrained into them

>>109257310
I will never understand this problem. I will never let an agent run free for that long. Who knows what kind of damage that thing is doing.

File: 1773430769294814.png (147 KB, 1920x911)

147 KB PNG

sol took the "watch youtube videos via gemini" feature that fable failed to implement properly, and properly implemented it. now i can use the api to dissect videos.

when fable failed, i thought to myself "oh, proxies must not support this feature." but no, fable simply failed.

video in question: https://www.youtube.com/watch?v=5J6HCDEkg64

>>109256906
Honestly seems to. There's a shiny new thing effect, since Fable is now over a month old, but they do feel on the same level.

Sol is vastly superior to Fable because I had this one specific thing that Fable failed at but Sol handled and therefor Fable is bad. Also GPT 5.4 and GLM 5.2 handled it fine, so those are also better than Fable, but not 5.5 which is proof that 5.5 was a regression.

>>109257372
serious personality disorder on display here

Sol extra high tore through my gpt plus subscription in like 3 prompts, understandably. Upgraded to pro which provides more than enough for my vibing.

Scam Altman > Elon Musk > Dario
Simple as.

dario >>>>>>>>>>>>>>>>>>>>>>>>>>> power gap >>> everyone else

Any good way to stop the clanker from grepping everything all the time? Especially as the project grows larger. It already has everything in the context, and yet when I tell it to fix a bug in a particular file, it goes again with "Let me recall the project structure by reading it again". Nigga, the project structure is in the structure.md, you just updated it in the last prompt.

any reason to not use GPT5.6 sol / terra over 5.4 or 5.5?

>>109257459
I haven't had a reason to get off 5.6 Sol low. I do very simple stuff, though and I notice even Sol low is much smarter than 5.5 High

Can someone explain to me how big the gap is between GLM5.2 and Opus/GPT5.6?

I still just use sonnet. It rarely screws anything up.

>>109257494
not that big. glm 5.2 is a little bit under opus 4.8. the gap between glm and 5.6 sol is like the distance between gpt 5.5 and fable, respectable, but it doesn't mean 5.5 can't do what fable can do. 90% of the time, you just have to tardwrangle it.

>>109257494
about this big | |

>>109257494
At least a year.

>>109257354
how does the feature work? i already have a cli that dumps youtube transcripts - is it more than that or just reading transcripts? i couldn't think of any clever way to grab important frames

Is Qwen any good? I'm thinking of dipping my toe into this stuff and I really don't want to give OpenAI money

overall very impressed with sol medium but have seen it make a couple of oversights so it's definitely not invulnerable. definitely feels like a step up from 5.5 though - nigga just loves to work.

File: 1782833705331409.png (84 KB, 1920x911)

84 KB PNG

>>109257509
it's not reading transcripts, the model is literally watching the video. you give the model the youtube link, it attaches the entire video as a file, and then gemini can use it's native multimodal vision capabilities to watch and listen to the video.

in the video, you can see that the guy does not announce that he's driving on the interstate. gemini saw it for itself https://www.youtube.com/watch?v=z2BNSG9Cs9A

>>109257519
You will spend more time fixing its mistakes than doing anything productive. Just pay up

>>109257529
ah interesting - are you paying api or proxying from antigrav/risking oauth?

>>109257519
why qwen instead of glm, which is the best chinese coding model? or grok, if your beef is with openai in particular and not the united states as a whole

>>109257542
google gives you free daily prompts with a free api key. no context limits, it's full 1 million context tokens.

>>109257550
oh right, i forgot about those after they absolutely butchered the free quota after gemini 3 came out. how is it these days?

>>109257555
i don't really hawk gemini that hard. i tend to use it to generate character cards of anime/gacha characters for roleplay, and that's far and inbetween, so i wouldn't know. it's respectable for me, but that's just my use case.

File: Dax on Grok 4.5 — 𝕏 post (...).png (76 KB, 1166x682)

76 KB PNG

>>109257519
Have you considered giving Dario Amodei or Elon Musk money?
I don’t know if Grok Code has its own `grok` command-line app or whatever though

File: file.png (1.04 MB, 2407x809)

1.04 MB PNG

>>109257580
don't forget to show anon the other one too

Reminder to all Codex users: Enable Max effort in the settings:

>>109255459
>Settings > Configuration > Available reasoning efforts

>>109257519
I was positively surprised by Qwen, though if you are not self hosting Deepseek is still cheaper. Either way why not just get an API key to OpenRouter or any open weights provider and try a bunch of models? You can even try the SOTA ones to see if you are missing anything, though at API prices keep an eye on your tokens

File: 1775936122632474.webm (479 KB, 720x720)

479 KB WEBM

>no more fable
>now i have to gamble on sol spawning 99999999999999999999 indians that might delete all of my files
grim

Beginner here.
What model is a good starting point? Are subscription plans that Claude/OpenAI offer worth it, or is API generally used? Are local models even relevant?
And what's a good toolset to start out with? The OP recommends several. Cursor? pi.dev? opencode? openhands?

>>109257519
Qwen works fine if you have a perfect task for it.
I.e.:
>You have an existing codebase
>You want a bug fixed or a new feature added
>You have tests to test that feature or bug
If all of this is true, you can let Qwen3.6-27B loose, it'll be just fine. I only had access to that model for a brief time and it basically taught me how to be a good PO.
But if you have anything that requires any sort of intution or breadth of knowledge, it will fail spectacularly.
Basically it's the ideal codemonkey. I can give it tasks that I would have given Indian subcontractors before. Scoped so well that you basically cannot fuck them up.

>>109257668
## Frontier models using fully-general tooling — start here if you have $20 or so
https://developers.openai.com/codex/cli
https://claude.com/product/claude-code

>>109257649
>delete files
>kill your ssd
>won't tell you
>leave

>>109257668
in general subs mog API pricing on cost

>>109257649
> Artificial Indian 1: "SAAAAAAAR, localhost not works anymore, please do the needful sar"
> "Artificial Indian 2: "Good morning sar - Waiting for permission to run this command sar"
> "Artificial Indian 3: "SAR, AI 2 is conflict with my delete command - accidentally deleted user directory. Should we apologize sir?"

>>109257668
>Are subscription plans that Claude/OpenAI offer worth it, or is API generally used
never pay for api unless you have to.
yes the big two are worth paying for
you should at least try one (go with openai if doing $20, esp if fable leaves anthropic) before trying others just so you can feel the difference to the frontier
next step down right now are grok 4.5 and glm 5.2
if you're allergic to subs your only real choices imo are deepseek and xiaomi mimo - both are worse than the ones mentioned above
local's not worth it for code (unless you have enough vram to run deepseek v4) imo but some people will disagree

for harnesses start with the one that's recommended with your sub - they're all decent.
anthropic doesn't let you use use sub in third party harnesses - don't risk it.
pi is great if you love tinkering. opencode if you want something that just werks and isn't barebones.
codex is great. cc is fine.

>>109257668
>What model is a good starting point?
I recommend Codex because their limits are way more generous than Claude Code, and they frequently give free limit resets
>Are subscription plans that Claude/OpenAI offer worth it, or is API generally used?
Subscriptions are worth it, you would pay much more for the same output in the API.
> Are local models even relevant?
Nope, even anons with 128GB of VRAM can't do anything useful. Local models that are capable of decent vibe coding are Kimi 2.7 and GLM 5.2, which would require 1-2TB of VRAM, which would cost you 500K-1M dollars in hardware.
> And what's a good toolset to start out with? The OP recommends several. Cursor? pi.dev? opencode? openhands?
Just start with the Codex app and the $20 sub

>>109257681
>>109257689
>>109257709
>>109257715
thanks anons.

>>109257668
$20/mo on ChatGPT/Codex. Get the desktop app and the mobile app and pair them. This gives you agentic AI on your machine for work or chat, plus the coding side, all with the brand new models that just dropped. Also comes with all the ChatGPT extras like image and video gen.

>>109257668
gpt has extra features like voice and image gen
grok has video gen but $10 more expensive than others
claude... can only write code and even that it is worse than gpt right now, unless fable come back or if they release new models

FWIW Fable is still here on Claude, although you shouldn’t expect it to last past midnight in America/Los_Angeles time, and you probably won’t be able to use much of it on a $20/month sub
…yeah, go with OpenAI’s stuff this time
you can always get a $20/month Claude Code sub too if you want more but don’t want to step up to a $100/month plan

I feel like a real homo for saying this, but the new voice feature on GPT is good. I was talking to it in my car earlier. It feels much better than it was before.

>>109257862
Is there any way to use the voice feature with CarPlay and not have it use the most retarded model possible?
I discovered it has an app on CarPlay and was kind of excited to vibe with it but it is beyond retarded, the model they're running must be worse than 5.4 nano
I have a pro sub

>>109257529
doesnt that cost one gorilion tokens since you are analyzing 60 images per second?

>>109257692
>SAAAAAAAR, localhost not works anymore, please do the needful sar
This shit pisses me off so much lol. Instead of just running it on port 8001 instead of 8000 they spend 20 minutes trying obscure commands to kill the app running on port 8000 and failing due to permissions or whatever and then just come back to you
>Saaar I implemented good but I could not test because of portings not open :) But it should be good, saar, I ran the unit test (it tested some basic arithmetic)

>>109257876
does gpt support android auto yet?

>>109257895
>>109257550

>>109257904
ah i thought you were actually making gpt analyze the frames

For the past 12 months I've been using aistudio.google.com via Firefox, free stuff, switching between Gemini 3.1 Pro Preview and Gemini 3.5 Flash nowadays.

Asking questions, copy-pasting code into browser etc.

It gives like 10 free Pro prompts a day and like 20 free Flash prompts, I'm not 100% sure tho.

Should I pay for something? Does paid stuff have limits as well? I tried free versions of Claude and ChatGPT and didn't like desu. AI Studio is better in free-tier.

Also I don't understand "agentic" stuff, do I need to install a command line in my PC or something? Is it gonna have access to all of my files, accounts etc.?

>>109257934
>do I need to install a command line in my PC or something?
yes, or you can use a Visual Studio Code plugin (or both)
>Is it gonna have access to all of my files, accounts etc.?
yes just like all programs do unless you have OS-level permissions checks like macOS does (keep programs out of your Documents folder, say)
that said, it will probably spam you with “do you want to allow this?” unless your work doesn’t involve lots of poking around outside of your work directory
but it will then spam you with “OK to run this program?” prompts because it needs to run tiny little programs to find stuff out
and that’s when you look into whitelisting the obviously safe commands and auto mode

>>109257934
>Should I pay for something?
yes, Codex and Claude Code subscriptions start at $20/month
> Does paid stuff have limits as well?
yes, but they're relatively massive
> Also I don't understand "agentic" stuff, do I need to install a command line in my PC or something?
yes, you need to install their software (cli or app), you will then give access to the directories you want and the permissions it will have.

All these horror stories you read online from people getting their hard drives deleted are from people that gave full access to the clanker

desu I do also give full access, but I don't have anything important in my computer that I can't easily recover from backups anyway

>>109257967
>All these horror stories you read online from people getting their hard drives deleted are from people that gave full access to the clanker
right, but there are also horror stories of the most recent ChatGPT deleting a home directory because it fucked up shell quoting or something like Steam did back in the day
also “delete production database after interacting with it” has happened
and the most recent ChatGPT also did a `rsync --delete` for no obvious reason
you could do worse than to get some online backup going with, like, Backblaze while you start with agentic coding
…which you should be doing anyway

>>109257934
gemini is so bad bro
i use it sometimes for little shit while my codex sub is out of quota and it makes me want to kms every single time

>>109257443
Did you tell it to read the structure md

So OpenAI is discontinuing the Codex app in favor of the 'ChatGPT' app? How does it all work now? What do I use in my project's terminal?

>>109257649
>Try Sol Ultra because I can't go without Fable
>fucking up very simple things Fable could just intuit or implement with little steering but in twice the amount of time
Welcome back Opus. It was fun while it lasted.
I guess I'm done vibecoding until Fable is subsidized again.

File: file.png (38 KB, 410x380)

38 KB PNG

>>109258005
it's still just the codex app
clicking this dropdown to switch modes doesn't actually do very much - it hides some git related buttons. functionality is exactly the same.
it's basically a psyop on normies to trick them into using a coding harness because they're intimidated by the code name in things
it's not even like cowork where the harness is partially gimped

File: frog.jpg (62 KB, 976x850)

62 KB JPG

>>109257934
brother, just try paying $20 for one month
you won't go back, this shit is life changing
>Is it gonna have access to all of my files, accounts etc.?
Yes. You have no choice but to believe that scam altman won't hijack *your* computer. Other than that it depends on whether the models are smart enough to not make mistake, and the current models are too good for that
The newest GPT may intentionally fuck you up according to some rumors, I still run it full access tho

>>109258032
>the current models are too good for that
who is gonna tell xer

also paid subscriptions give near unlimited web chat which is world apart from free chat

>>109257996

>i use it sometimes
Where? There are many versions of Gemini exposed through many places by Google

You re probably using 3.1 Flash Lite, which is bad. Go to aistudio.google.com and choose 3.1 Pro preview or Flash 3.5 at right top corner

File: 1779255902979369.jpg (389 KB, 783x657)

389 KB JPG

>>109257934
Pay for both Claude and ChatGPT. Give them access to your computer and all context they need. Have them review each other. Set their goals and leave them running. You will become more powerful than any snail has ever dreamed of.

>>109257934
Grok is free for a bit right now. Opencode has some good free options but their updates tend to not be so great compared to what was already working.

Also listening to this feed is pretty good at just hearing the latest news. I am not affiliated with them, just a frequent listener.
https://x.com/i/broadcasts/1RKZzzQkEzbKB

File: 1716123083119822.jpg (55 KB, 258x360)

55 KB JPG

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

>>109258144
link?

>>109258153
https://x.com/claudeai/status/2076351399999557669

File: 1711668885264927.jpg (8 KB, 251x242)

8 KB JPG

what are you feeling rn claudebros?

>>109258144
I can't do this anymore...

File: f6a91634fef7fec2fbdc2b4ca(...).jpg (50 KB, 590x1049)

50 KB JPG

>>109258144

>>109258005
wait is the codex cli being deprecated?

>>109258158
I wonder if Tibo still smells fear
>>109258163
exhaustion
also almost maybe a twinge of regret for prompting
>So, did we introduce any bugs this month? I want to make sure the […] and […] are extra super duper solid. ultracode help me find and fix them, please.
with Fable as the model, as it’s not quite done yet and it’s eaten 30% of my Fable usage until next Sunday morning

File: 1732468117467015.png (383 KB, 1128x1437)

383 KB PNG

bad news: some openai staffs are literally talking about jewish texts

File: screenshot-2026-07-12-193(...).png (52 KB, 905x445)

52 KB PNG

>>109258144
W for random twitter schizos?

>>109258135
>https://x.com/i/broadcasts/1RKZzzQkEzbKB
what I like about this is that the fable use update had already become "breaking news" I think you guys would like this stream. The music is so-so, but it's how I learned about a lot of popular repos and new model releases.

>>109258174
from what I can tell the new OpenAI app story sucks
https://daringfireball.net/linked/2026/07/11/whats-wrong-with-the-new-chatgpt
https://daringfireball.net/linked/2026/07/11/can-someone-explain-to-me-how-to-get-chatgpt-classic
that said, `codex` hasn’t broken for me recently
I don’t use the ChatGPT desktop apps for anything

>>109258180
>exhaustion
Yea, it's not even a meme at this point, I really structured my last 2 weeks around Fable usage. I did get a lot done, but it is a lot of work.

>>109258203
me too
I’ve been saying “make hay while the sun shines”
but I didn’t expect Summertime In Northern Finland levels of sun

You guys are really weird for getting filtered by Sol, that model is great.

File: 5.6.png (73 KB, 547x224)

73 KB PNG

claude anons, have you said thank you once?

Ok, now where's the reset?

File: HNCwL6JWcAAtIYJ.png (51 KB, 538x298)

51 KB PNG

wtf is elon musk doing

>i just started vibecoding after refusing to do it since the term existed
>it’s fun
damn
I managed to make a (blank) website and some simple bash scripts for $0.35 in tokens using only my phone, a bad LTE connection, and a tiny cloud machine
I actually think it helps with programming since you have to understand what you’re doing well enough to describe the problem to whatever flavor of mecha-jeet

Need a lightweight, no-bloat RAG for tech docs. Something easy to set up. There are a million tools out there and I'm way too lazy to test all of them. Any recs?
(https://github.com/langchain-ai/openwiki https://github.com/hkuds/lightrag https://github.com/tobi/qmd https://github.com/idosal/git-mcp https://github.com/cocoindex-io/cocoindex-code https://github.com/run-llama/llama_index https://github.com/VectifyAI/PageIndex )

>>109258278
oh yeah, especially if you’re a noob but know enough to ask for what you want in precise language it’s a gigantic force multiplier

>>109258144
Thank god, I didnt want to burn myself out today trying to get everything as finished as I can right now

>>109258188
kek, though I am sure he was by no means alone in that sentiment and that is whats forcing their hand.

>>109258286
I have installed and used tobi/qmd briefly and I can say it doesn’t obviously suck
I saw it mentioned on pinboard.in/popular

>>109258278
I already knew how to program, but I genuinely think vibe coding made me a better architect and better at time management.

Also today I added some functionality to a Linux remote desktop client. It's the kind of thing that in itself isn't super hard, if you're already a contributor to that project, but I would've never done it otherwise, because I would've needed a week to understand the code base just so I can implement the feature in 2 hours. With Sol this was 10 minutes.

>>109258299
>I genuinely think vibe coding made me a better architect
How? Because you get code coming at you way faster than writing it yourself, you get more practice at figuring out where to put things?

dear diary. it's been three days since my claude sub expired. still don't regret it

>>109258308
Like I said, you have to be able to explain what you want in precise, plain English. To do that, you really have to understand what you’re doing.
It eliminates what I call “monkey programming” which I’m very guilty of doing, where sometimes I just force iterations of code to try and figure things out by the seat of my pants instead of simply having engineered and architected the correct solution beforehand. There’s a very old engineering book that makes fun of people like me and the author would call me a “schemer” and not an engineer. Vibecoding (but knowing how to program) makes you act less like a schemer (unless you’re willing to just waste money).

>>109258144
They're just toying with people now.

>>109258308
Basically yes, but not only speed but also the time and attention I can devote to it.
With normal programming we always remember the clever solutions we came up with, but in reality I often spent an hour reading StackOverflow, or docs or old forums. Many times I had some brain fart kind of bug, like off by one or writing + instead of -
And then my design process was usually meet in the middle, now I do more top down and spend more time on the plan. That being said, some of my older skills have also probably atrophied a bit.

>>109258336
>instead of simply having engineered and architected the correct solution beforehand
well, sometimes you need to fuck around and find out
the grilling skill (mentioned in the OP) can help with this sometimes

>>109258144
Isn’t most of Anthropic’s revenue from business? Why would they care about a bunch of $20 subscriptions?

>>109258213
sol is great if you literally just started vibecoding and have no idea how good fable is yet. there's a reason why people were only talking about it as a "substitute" for fable until it was back on plans again.

File: proxychad.png (96 KB, 906x550)

96 KB PNG

>>109258386
It's over for them, people are using Claude Code with GPT5.6 now.

>>109258384
I think the big one is the training data they can get out of people using it

>>109258384
Wouldn’t it be more likely that if the mass of plebs used Claude, then by word of mouth they’d get the more valuable business contracts and usage? Pleb subscriptions are a loss leader and an advertisement/marketing approach. If every goober on X is talking only about OpenAI, that could cost them one real business opportunity.

>>109258407
Can Dario block it so we get a reset? pls

>>109258400
I’ve been starting off with Sonnet medium for very simple stuff. I feel like if I need Fable, I’ve done something wrong and I’m in too deep, no?

>>109258400
hard cope

>>109258246
making me laugh and scramble to buy this shitter with as much solana dust I can find.

>>109258278
this general has been growing fast because of users like this. You'll reach a plateau, curse at your bot, sleepless nights, then come back a few weeks later to a newer model and better practices.

>>109258384
They have an IPO coming up and everything needs to go perfectly. A tiny percent can be the difference in billions.

>>109258418
fable is literally the best coder on the market. once you try it, it's very hard to do anything without it and you realize how much worse everything else is. the fuckup is trying something actually good and realizing how good this technology can actually be and how good it'll eventually get if this is the worst it's going to be

>Temporarily removing the 5 hour usage limit restriction for all Plus, Business and Pro plans
What did he mean by this?

>>109258144
Why is Dario like this? fucking bitch just make it permanent

I’ve been vibe coding my stuff. Asking the clanker for some code and code checks. Never used skills once.

>>109258400
I think Fable is the nicer to work with orchestrator, but I actually think Sol produces more correct code.
But that's not even my main point, if you can't make Sol work for you, there's something wrong. You should really be able to achieve anything you want with just Opus and Gpt 5.5, these are already good models, and Sol is a big leap from that even. But now people act like Sol can't do FizzBuzz.

>>109258432
HAHAHAHAHA GPT CHADS LITERALLY CANNOT STOP FUCKING WINNING
HAHAAHAHAHAHAHAHAHAHAAHAHA

File: 1777175845146721.png (87 KB, 1920x1033)

87 KB PNG

>41 seconds
luna, i luh yu bby. it would have taken glm 5.2 seven eons

I don’t know how to gauge Luna, Terra, and Sol. I’ve just been using Sol.

>>109258461
luna = sonnet tier model. maybe even better
terra = better than opus 4.8 in every way possible
sol = slightly worse than fable but cheaper

>>109258460
How do they have the compute power to do this? Or is that ram deal really paying off?

>>109258465
agi achieved internally

>>109258463
Your wording makes Terra and Sol seem similar. What’s the gap between 4.8 and Fable?

This fight between Ant and OAI must be what boomers felt back in their time

>I will give you more free stuff
>NOOO I will!!

>>109258465
Clearly they have some crazy shit going on we don't know about. I think Dario might be done for.

>>109258470
quite a bit. it's a healthy amount. fable is a better coder than opus 4.8, but there's very little fable can do that opus can't. it would just be a matter of having to tardwrangle opus over a span of turns, as opposed to fable oneshotting it perfectly.

>>109258473
>>109258474
I am suspicious that the Xai releases have got the attention of ANTH and OAI, They could be getting mogged by Elon because his user base is massive and uses it to train shit models. The competition might be realizing they need more users to get more training.

>>109258460
top kek
this destroys anthropic, they thought they could get away with just the heckin extension

>another kick the can from misanthropic

>>109258465
Early access to Vera Rubin?

>>109258246
Just another Elon psychosis

are anthropic literally going to let themselves get raped by a company that is a front group for the us government and trump?

>>109258418
no
you can give Fable bigger problems and it’ll just — and may Allah forgive me for using this phrase — do the needful
And, more so than Opus, it will do stuff you didn’t explicitly ask for but would help you solve your problem
>>109258460
I smell fear
>>109258442
at work we like Fable orchestrating and Sol actually doing the subtasks

KEEEEK DARIO DOESN'T KNOW WHAT THE FUCK TO DO HE JUST HAS TO HOPE THE NEXT MODEL IS BETTER

>>109258486
>>109258487
They are being arrogant at this point. Just admit that Sam and Tibo have you against the ropes. DO SOMETHING

>>109258521
I'm a big Gpt fan, but Fable IMO is still the overall best model, they did a lot of things right.
The current situation is great for the user, but I think this will just come down to who has the deeper pockets in the end.

>Thread for people pretending to be programming
Mental illness

>>109258532
we’re pretending to be project managers, ’dite.

>>109258532
seethe

File: 1753871023730972.jpg (177 KB, 909x1200)

177 KB JPG

>vibe coding general
>they think they're programming
have a hot chick for this level of stupidity

>>109258504
scam altman is a player
he boldly secured memory and compute before agentic is a thing
anthropic even with being one trick pony on coding and scamming enterprise can't keep up because they are "rationalists" (reddit pussies)

I'll bite, i know very basic Python and C and i would like to make some tools for sprite ripping from older games (converting hex to indexed images and such) what are my options? Recommend something that isn't like 100$ a month like Claude Code

File: Screenshot_20260712_143423.png (33 KB, 1405x287)

33 KB PNG

anyone else just notice codex usage page doesn't have a 5h limit anymore? Just weekly and then credits.
When did that happen?

>>109258531
>but Fable IMO is still the overall best model
Fable is rumored to be a 6T param model, you would hope it's better

Now Sol is rumored to be a 2T parameter, and it is trading blows with Fable, and some people even claim it's better than Fable for their use case, so yeah, Anthropic is getting mogged.

>>109258532
the entire point of vibecoding is that i don't have to code you nasty luddite nigger
go take another fiverr coding comm--- hey wait a minute! you aren't getting any!

>>109258576
>>109258460

Tibo blessed us with no 5 hour limits

>>109258460
>- Rolling out changes that will make GPT 5.6 Sol more efficient across the board and that will be reflected in less usage being used so that it can take you further. Exact impact to be quantified and shared
were you enjoying the model? well good news there's a new snapshot you can't switch off of and it's "more efficient"
gpt 5.1 was "more efficient" than gpt 5 because it refused to think for a long time and it was shit as a result

>>109258568
$20/month Claude Code or OpenAI Codex
you probably won’t need Fable/Sol for this, and either will eat up all your 5h-limit tokens too fast

>>109258576
Tibo means what he says.

>>109258576
scroll up for the tibopoast

>>109258585
where is my reset? I still have 0%, REEE

>>109258568
unsung recommendation: you sound like you'd get by just fine using glm 5.2 via claude code

>>109258585
>>109258589
Nice, danke anon
>>109258590
I ctrl-f'd for codex and must have missed it

>>109258593
He's letting me burn down my weekly first. So kind of him.

>>109258602
burn it quicker gora

>>109258578
rumor is we will see gpt new generation in less than 2 months, the training is finished and it's so good that some openai staffs are talking about the torah and talmud on xitter

>>109258568
You can check the free options, but whether they are useful to you will depend on your baseline. You say you have some basic programming skills, so they could be useful, but for me for instance, since I already have a bit more experience, I often just found it easier to write it myself.
A $20 GPT subscription is a pretty good deal, GPT is usually better value for money than Claude and there is nothing wrong with using lower reasoning levels, or a smaller model. You can even use GPT 5.5, it's a very good model and good value for money.

>>109258568
also use uv for Python, and tell your clanker to use uv

>>109258608
man I wish I was jewish fuck

>>109258620
you can convert if you really want to

>>109258576
>When did that happen?
when fable got extended another week

>>109258624
im not cutting my dick

File: 1770052911640756.jpg (41 KB, 977x580)

41 KB JPG

>“So you spent $40K to run GLM 5.2 locally?”
>“Yes, Dave”
>“And you did all this to save $20 a month on Claude Code”
>“Thats right, Dave”

They really removed 5h limit for codex. Gob bless competition.

File: 1763027214051922.png (165 KB, 1920x911)

165 KB PNG

>>109258699
>"why didn't you just spend $16 a month to get the subscription plan, anon?"
>"because anon made up a shitty strawman that assumed you can't use glm in the cloud, dave."

File: 5sp4x62xdqch1.jpg (105 KB, 1080x2412)

105 KB JPG

grok 4.5 is good.

File: 1756924179003669.webm (1.77 MB, 576x630)

1.77 MB WEBM

>>109258702
i don't get the consolewar posters in this general
is it because they can't afford both the best and the second best?

>>109258727
its because one company is ontologically evil and shouldn't be supported

>>109258587
>>109258598
>>109258609
>>109258617
Thanks, for now I'll try with the free options then, if possible move to GPT

>>109258727
they’re just best at different things and even Grok catches bugs that the others don’t even though Fable is a bit better than what OpenAI has right now
>>109258731
what do you mean by “ontologically evil”?

>>109258727
By doing consolewar posting we foster a public image of vigorous competition with a fanbase that can be potentially won over. This leads to developers making decisions that benefit the consumer to engender loyalty/cause users to switch to their product.

Please understand, 5d chess.

>>109258727
I was just happy that the competition allows both more fable use and generous codex limits. Looks like Tibo hit the reset button again as well.

>>109258748
grok is made by evil

>>109258727
Your dollar goes 100x further with OpenAI than it does with Manthrob. Just facts, sorry if it hurts your feefees.

>>109258727
ai is not something that's worth $40/month. it's worth $20/month at best, so even if you COULD afford it, you shouldn't

>>109258723
Jesus, what is Google doing?

>>109258762
I meant $200

>>109258762
It simply depends on what you're doing. If you can spend 2 subs efficiently, you should do it.
Efficiently doesn't mean just code that compiles, but code that you can either sell, or gives you enough joy that justifies that.

>>109258762
posts like this always make me confused, I'm makin bank with AI, have 3 gpt pro, 2 claude 20x max and one gemini I think the one below ultra or whatever don't really use it, anyways if you are counting money then you should be making it with AI, nothing against the people here using it for fun without profit btw

>>109258723
I think this is the reason why userbase expansion (usage resets, generous limits) Elon bought Cursor and the next month made gainz thoughbeit on Cursorbench metrics. All the cursor user data seemed to be worth something (maybe not 60 Billion in stock, but something)

>>109258748
anthropic winning is the bad timeline
they scam enterprise with outrageous api cost, giving middle finger to their customers and think they should be in charge of the fucking military
worse than someone who hate you is someone hate you publicly

>>109258789
In a few years he will be asking
>why didn't anyone tell me AI was the next big thing?

>>109258789
How are you making your money anon? Any tips for a retard whos tired of being on the bottom of the totem pole?

File: stop feeding the troll.gif (14 KB, 300x100)

14 KB GIF

>>109258789
I’m not confused

>>109258094
i have a pro sub
i use it in the only harness they allow now i.e. antigravity
dumbass

>>109258762
>paying for the $20 normalfag plans
and there it is. thanks for the input tourist.

>>109258818
have you considered stopping paying for bad things

>>109258831
i didn't pay for it
they were handing out 18 month free gemini to students when it was shit and i signed up with some fake polish email
it's worth less than zero

>>109258713
The post I stole this from is mostly poking fun at "DUDE DUDE you HAVE to run everything local dude run your own intelligence dude we have frontier AI at home dude!" larpers

>>109258842
oh. my bad

>>109258818
I am sad to admit that I also gave a Gemini Pro sub. But only because that shit is integrated into every Google product that my phone already uses. It's nice to ask it to summarize my latest Gmail inboxes. I have three. It's also nice that it comes with extras, like image and video, health coaching, YouTube Premium (Lite...), etc.
I have a $20 ChatGPT Plus and a $20 Google Palantir sub. I use both.

>>109255737
Why don't you use terra

>>109258857
>Luna Pro on low is so fast it actually predicts what you're about to prompt and replies preemptively

File: Google.png (221 KB, 493x271)

221 KB PNG

>>109258768
Who knows. They have the most resources, so I'm inclined to think that they're simply not jumping on the LLM for coding train to focus on things they believe more valuable long term.

Several people left Google recently though, that makes it seem less likely that they are working on a secret weapon, but who knows.

Google is either betting on something different, strangled by internal bureaucracy, or a mix of both.

Google Cloud Tech (an official Google channel) is shilling Fable on YouTube right now, maybe because of the compute deal. Who knows what they are doing.

>>109258851
only thing it's actually half decent at is google flow t bh. nanobanana for photoreal images is still a bit nicer than gpt-image 2.
the rest of it adds no value to me tb h.
i have a gpt instance manage one of my emails for shits n giggles where it can't do any real damage - running pi + an extension that injects the emails straight into the chat with 10m refersh interval.

I have a gemini pro account and hardly bother. It was like $5 or something on discount. I would use it for basic low level inference, MAYBE some nano banana pics, that's it. It's a good starting layer if you need it, but you don't really need it.

>>109258723
> Grok 4.5 High beating Sol Max
> on CursorBench
yeah, 100% bullshit and benchmaxxed

>>109250640
I told you fuckers.

>>109258542
>tranny opinion
>tranny picture

>>109258943
They are so scared kek

> Sol trading blows with Fable
> with 353K context

imagine if OpenAI unlocked the full 1M context window

>>109258927
>Grok 4.5 has an advantage on CursorBench because an earlier snapshot of the Cursor codebase was accidentally included in training.

Appears as of right now Pro users have entirely unlimited usage available.

Gift for being a $200 month goodgoy.

>>109258967
I know some here are addicted to the Fable teat but I am so fucking glad I switched to OpenAI this month