A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## News- (2026-07-09) OpenAI Sol/Terra/Luna released- (2026-07-08) Grok 4.5 out- don’t ask me, I don’t keep track of the news## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/clihttps://claude.com/product/claude-code## Not worth it for code, but maybe good for other thingshttps://geminicli.com/docs/https://x.ai/cli## Open / local / self-hostedhttps://github.com/OpenHands/OpenHandshttps://github.com/QwenLM/qwen-codehttps://github.com/QwenLM/Qwen3-Coderhttps://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF----## Prompting / context / skillshttps://arps18.github.io/posts/claude-code-mastery/https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://github.com/mattpocock/skills — /grilling is a favorite## Other editors / terminal agents / coding agentshttps://pi.dev/https://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/https://docs.cline.bot/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent## UI/Frontendhttps://www.figma.com/make/https://www.anthropic.com/news/claude-design-anthropic-labshttps://uiverse.io/https://ui-ux-pro-max-skill.nextlevelbuilder.io/https://stitch.withgoogle.com/## In-browser builders / hosted vibe toolshttps://bolt.new/https://replit.com/https://docs.github.com/en/copilot/tutorials/sparkhttps://v0.app/docs## Benchmarks / rankingshttps://www.tbench.ai/leaderboard/terminal-bench/2.0## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109226674
i freed factory.ai and thought it'd be a waste to keep it private, so im releasing hereyou need a little braincells to ask your aislop to analyze the code for malware though, and compile from source yourself (never trust binaries)its a great coding agent overall but i wouldnt pay $20 per seat monthly for this (you still pay even if you use byok)so enjoy i guesshttps://github.com/udonge-foundation/drool
>>109232574>hacker known as json
>>109231917>and don't think it's not as good>implies he did think it's not as goodESLs should go through an LLM before posting online
>>109232619nvm it's a jeet
>>109232574that's how you make your boss happy. tokens = performance.
>>109232642only when you’re going from noob to midhttps://steve-yegge.medium.com/the-flat-curve-society-36c8b01eb33b
>>109232619This is just a normal editing mistake, I generally can't be bothered proofreading for social media any more. Reading back I notice I've ommittsd a word or failed to clean up an dit about 20% of the time, but unless it significantly alters the meaning it seems pointless to put time into fixing it.
>>109232666
Genuine question:Do tools like ponytail, caveman and codegraph actually do something?Ponytail I've tested myself but I want to know what other anons think.pic is my first finished vibe coding project (out of 6 started)
>>109232709
>>109232712I don't believe in prompt engineering
>>109232827then you are only wasting time doing fuckall
Training a recurrent adapter to turn a regular model into a latent CoT model by converting output activations to input embeddings in a differentiable way.If this works I want to backprop through the whole CoT to optimize it without RL.
Working on TumblienAbout to ship this, but still lots to do before it's properly released. lots of features planned, etc.
>>109232860>doing fuckallisn't that the essence of vibecoding? doing "something" that is actually nothing, is useless, waste of time, money and produces 0 income?>inb4 we trillionaries nowhahahahahahahahahahahahahahahah
>>109232900Money is not what defines worth to all people. You are not on /biz/. We are on /g/. It's okay to do things for fun.
>>109232712ponytail actually does, from what I've seen so far. It cuts out a ton of bloat and writes way more efficient code, so it's naturally more token-efficient too. What I really like is that it ONLY does the thing I asked for and doesn't randomly tack on a bunch of unrequested bullshit. I didn't install the whole setup though, just copied the skill into a ruleset for the project (.claude/rules/) or dumped it into CLAUDE.md, since that's basically the core of it anyway
>>109232906cool story nigger, I get paid to laugh at ai code and discard it
>>109232610preddy cool. are missions the only real hook they have?
>>109232712>Do tools like ponytail, caveman and codegraph actually do something?yes. they degrade model performance.great at making midwits feel smarter tho.
>>109232920This is how I can tell you don't have a job
>>109232925>ponytail, caveman and codegraphI've never heard of any of this He also took a windows screenshot, I'll hazard a guess that this is windows tooling? I'll look at it, don't expect to find much though
>>109232934there are a bunch of these things out there - most are just skills telling the model to talk a certain waythe worst ones actually intercept the results of common tool calls and rewrite themall this as a sad attempt to decrease token usage, not understanding that most tokens are invisible to the userlike i said, it makes midwits feel smart
>>109232944I gave one a shot for now, they're agnostic to the OS platform which is a surprise to me.I know MCP, Agents, Skills, Guardrails, Workflows already. Am I missing anything?>Tool callsAnyone who charges by tool call invocation is fucking scamming. Imagine if we charged $20 for every method call in a legacy Java framework.
>>109232955just talk to the clanker, bro.that's the only thing that will matter soon.all this other shit is going away.
robot gf when
>>109232973Stay out of /vcg/ if you don't like vibe coding
>>109233031i don't like psueds who think in 2 years you'll still be injecting skill file frontmatter into the system prompt of a god
>>109233037during the luddite purge you will be first in line
is that local Qwen 3.6 from sticky worth bothering with? can it be used for any real work?
>>109233114your reading comprehension is poor so i will no longer continue this bit
>>109232827Reality does not care about your beliefs.
>>109232827you ludlost
>>109233037sure anon, the model will always get it right and never get it wrong, you're the smart one here
>>109233214the model will get it wrong onceand then never againoh ye of little faith
Lots of low IQ people in this thread.
I'm not a luddite I just don't believe in value of generic prompts
>>109233232131. Low but never really bothered with coding so it's been a fun little thing to pass the time.
>>109233232Yeah they really misread what you wrote lmao. Probably third worlders desu.
>>109233229sure but why wait two years when I can just do it now
>>109233121Like old Claude would say, "it's not nothing".
"Remove ...">+1057/-20
>>109233232>vibe-coding>high IQ
>>109233282everybody does it, it's neither low nor high IQ.
>>109233311FUCK MAN WHY DID 4CHAN X SHIT MY RESPONSES UP.I was asking if someone "Have any anons been able to figure out how to jailbreak opus 4.8 or at least fable? Maybe Grok 4.5 or Gemini 3.1 Pro fuck even Sonnet 4.6 or Opus 4.7 ">>109233031surely his website already had traffic. i doubt fable can make a good business plan from scratch
Why isn't Gemini useful for coding? Is it at least better than local?
>>109233311what did you do
>>109233342its a meme i found in the jailbreaking discords
>>109233346maybe you want to participate in the official bug bounty program ethically and they just ban you https://hackerone.com/anthropic-cyber-jailbreak/
>>109233359No i just want to import these logic structures onto opus 4.8. Found them here https://x.com/Av1dlive/status/2074796427595874636I was using fable to reason maxx and then it said it switched to Opus because of muh ToS . it burnt all of my fable tokens and didn't even execute the task well.
this is like that kojima movie review meme
>>109233383to be fair every single one of us with money are eating absolutely good right now. which makes me ask ...>>109232712>>109232944>>109232955What MCP, Agents, Skills, Guardrails, and Workflows do you guys recommend for literature review and dissertation generation?
>>109232666Tldr, whats up with the pigs
>>109232666What a garbage article. This retard needs an editor to tell him to cut the crap and get to his retarded point quicker.
>>109232610Whoa
this one's a keeper i think.
>>109233501what are you doing with it anon
>>109233359you need credentials to be accepted for that, no?
>>109233383>it's just so fun
>>109233512i've saved it to my meme folder next to shiggydiggy.jpg
What time do you guys think the new model will drop at?
>>109233603Morning to noon California time zone.
gives you a sense of just how delayed this model really is.
>>109233632breh
>>109233632But 5.5 was released 2½ months ago...
I need my proxy to give me 5.6 already so the raep can start
>>109233651ASI was already achieved internally retard, they are just milking us before the big release in 27
>>109233666meds now
>>109233332It's specifically trying to be a good general purpose model. The Gemma team seems to be somewhat targeting the consumer market with their releases. Not many people can even run ~35B models like the Qwen Moe of the ~27B Gemma4 Moe, but many have enough for their recent 12B Moe and below. This is my anecdotal experience talking but for any vibecoding model to be good I don't expect anything decent from anything lower than the 24B range. The model's perimeter count heavily influences performance for my calculus says the Gemma 14 weight the options and decided going through the extra work to specifically train in good vibe coating performance had diminishing returns because the people that would 1) be capable of running one and 2) would care about that in the first place, likely already had and use proven models like either the Qwen models or some external api model with a bunch higher parameter count. Recent general purpose models are more than good enough for tool calling and local web search and don't necessarily need to be good at coding or "agentic tastes" to do those well. If local ai usage ever becomes a widespread thing (lol, lmao, thanks @sama) than most people will probably use the Google models because everyone knows what Gemini is and will probably just gravitate toward their shit anyway.
5.6 gonna suck dick, isn't it
>>109233692so what you are saying is that gemma 4 is the best model we have?
>>109233701Even 5.5 is still useful.
>>109233716Currently It's the best model family for vramlets, yes. Check their hugging face page. They have shit that is meant to be run on high-end gpus or shit that run on your average consumer shit rig
>>109233722by vramlet you mean below 6000 rtx?
>>109233679>>109233666>>109233651>>109233632One or two of you anons need meds. Theres something fishy going on with that timeline and I'm inclined to believe Tim over random anons
So what can i run with this? I want sonnet 5 locally. fuck even gemini flash with deep research
Anyone tried the new Grok? Is it actually good now or still a meme?
>>109233762i am trying it as we speak. its really fast. i am genuinely impressed with it. note that i also use opus 4.7, gemini 3.1 pro, opus 4.8, fable 5... Grok 4.5's ability to cache relevant context is mindboggling.
>>109233745how do I convince my boss to get me one of those?
>>109233782show him pic related
>>109233781How good is it compared to 4.8? I've heard that it's really token efficient, and if it's even anywhere near 4.8, I'll help fund Elon's second Trillion.
>>109233788so is GLM-5.2 equivalent to sonnet4 or opus8?
>>1092338034.8*
>>1092338024.8 is a lot slower that I know. Honestly I need to keep testing. I'll know by the end of the day and I'll end up posting about it here stay tuned.
>>109233737I believe him too, it's just weird that 5.6 would have been ready when 5.5 just came out. That's like someone saying they're using GPT 6.0 right now.
>>109233803100% its a case by case basis. >>109233817>That's like someone saying they're using GPT 6.0 right now.Fucking truth nuke.
i am becoming insane
>>109233803No, it's barely Opus 4.6, probably even worse. Extremely benchmaxxed for pretty graphs, can't do shit when you give it a real world task
>>109233788>just use worse models brotoplel
>>109233837You dont understand how to run a business kek You think walmart became what it is shipping the highest ever quality products?>but but look at this 1% exceptioncope>>109233831real world tasks arent generating ai porn for you anon
>>109233817I don't know who the guy is, but at least some OpenAI employees probably already are running some beta version of 6.0
>>109233745with 94k you could use sonnet every day 24/7 for a decade
>>109233877and be left behind
>>109233880As if that hardware won't be outdate in 3 years tops
>>109233823go with god anon
>>109233882anon that hardware for companies that are into this sort of thing can fire 3-4 people, buy the newest model in 3 years, and still generate 300-450% savings >>109233232A lot of low IQ peple in this thread
>>109233890Open source can't compete with API sota, if you company is using open source you should jump out as it is a sinking ship
>>109233745>>109233782You would only be able to run 1T models at Q3 and slowly.It only has 250GB of actual VRAM and 500 of meme unified memory.At that point get a Pro 6000 with DDR5 and it'll be probably about the same $60000 cheaper.
f-fable bros... i'm not feeling so good...
>>109233232and the mods lowest of the low
>>109233901im already using sol, I would say if you know prompt engineering it is just as good
>>109233890LMAO no.You ain't replacing anyone with that thing.At most it'd be a bad coding assistant for companies that NEED to have on prem AI or use cases that only require a dumb model. Other than it's dead in the water.
>>109233892>you should jump out as it is a sinking shipEveryone had your mentality about linux being a daily driver then valve invested and look at us now. investing in open source early on now and knowing the ins and outs as a legitimate ai research scientist (if even self-taught) will be the equivalent to holding bitcoin since 2011.
Is Perplexity usable for coding?
>>109233923I wish you were right but that's just wishful thinking. In reality the only thing that will matter is the capital you had when human labor became obsolete.
>>109233923You are mixing a bunch of stuff together.AI is cool.Open Source models are cool.A rich company using Open Source instead of SOTA API is a terrible move atm, might change in the future.If they are using local for labs and research then great, but if they are using it in their pipeline instead of Claude/GPT then they are shooting themselves in the foot.
>>109233928is youtube a good programming language?
>>109233892low IQ takehierarchies existhierarchies of tasks existif you're asking fable to fix a small bug or correct your css alignment then you're just dumbyou use sota models to architect and give directionopen source models, even seemingly retarded ones that are trained on coding, can execute the implementation plans, and your sota only reviews it.you save a lot of money this waythere a gazillion of tasks that can be done by open source modelsbut unfortunatelythe people are retarded
>>109233930You need to wake up anon. Human labor isnt becoming obsolete until 2050-2060. Until then the best you can do is find boomers in poor states who own businesses, and showcase them pre-made websites with an elevator pitch asking for a price for you to hand over a new company website they can use from then on out. You could do the same for thirdies who own companies.
>>109233944You have Opus, Sonnet, Mini, and Nano for that. I never said everything should be run on Fable.
i am once again encouraging people just clone existing services and sell them for a 1/5th of the price
>>109233953no, you said i have to use paid models to do things open source models can do for freethat's what you saidand you said it again on this messagebut unfortunatelythe people are retarded
>>109233948By pre-made obviously I mean you would have to do research on their business, create a highly curated site for their usecase as if you were already employed, blow their mind, and get paid easy. You could even sell at a market discount because youre making quantity fast anyways.>>109233956Not bad either. Thanks anonWAGMI
>>109233956>just make your own Fable bro
Do any of you have Hermes Agent set up?
Just use PI
>>109233964humans are just fable wrappers
>>109233964I made my own Fable already using Hermes.
>>109233964No tard hes talking about the shit on producthunt thats all vibecoded (and yet has funding) Like this https://www.producthunt.com/products/compendium-2
>>109233974Fuck I made my own version of that months ago. I really wish I had the mindset needed to do all this selling shit.
>>109233974I see, that makes more sense.
>>109233979Buy caffeine pills or get on modafinil which you can find really easily on X.
>>109233960Best opensource models are still months if not years behind the latest worst paid models. For roughly the same price.
>>109233991no they aren't, latest glm is 90% as good as fablethe people are retarded
>>109233991no they are notyou never used themyou need to double down on your retarded takebecause your brain cannot process the fact that you're wrongyou will grow upand luckilyyou will stop being retarded
>>109233762It is token-efficient as far as I can tell. Roughly as efficient as Composer 2.5.
>>109233974>>109233979>>109233988Not him but everybody and their dog has their own memory system for LLMs.From what I gather to succeed commercially at something like that you have to be an obnoxious person and shill your shit everywhere, lie and exaggerate, and have connections or LARP as some kind of "influencer".
see the thing about a smarter model is that it can just do things a dumber model can'tso you it doesn't matter how much cheaper your tokens are - the task will cost infinity because it'll never get doneor worse - it'll get done in a broken wayand you just won't knowdon't take the riskpay for high quality tokens
Is codex now censoring its thinking summaries? I am running 5.5 on opencode, though it is a version a couple of months out of date, but I just noticed thinking traces look like this now.
>>109234024>you have to be an obnoxious person and shill your shit everywhere, lie and exaggerate, and have connections or LARP as some kind of "influencer".man I wish I was born with psychopath genes, goddamit
>>109233999Is this actually true?
>>109234031probably just backend changes prepping for rollout - harnesses will just have to update to account for it
>>109233974>https://www.producthunt.comnever heard of thatetsy for vibe coders?
Fr tho is it even worth to not vibe code anymore? I don't code at my job anymore it feels like a waste of time and things still work. Perhaps making/mantaining shitty cruds apps doesn't need big knowledge?
>>109234045Ah well, shame I liked reading them but I am not updating opencode unless absolutely required, fuckers break something new every update
>>109234054do you live under a rock? what the actual fuck
>>109234060I don't know what circles you frequent but I never heard about it and I'm pretty deep into the LLM rabbithole.
>>109234068It is from way before AI.
>>109234068its been around way before ai. its aimed for startups though these days startups are just vibecoded
>>109232574>OpenAI Sol/Terra/Luna releasedwe really need three of these fucking things
>>109234114let them cook
>>109234114Your orchestrators need orchestrators. Agentic delegation. Looping prompting the prompt loops. Every step will have its own perfect model. Trust the plan.
5.6 when?
>>109233948I don't know man. I worked developing business process automations for a consultancy company and I've been laid off a month ago. And from what I'm hearing the market for anything dev or IT related has been very bad for a while.
>>109232574I ask chatGPT about code, review the result then paste it.Am I vibecoding?
I’ve been using gstack skills, and aside from some of the YC cringe stuff it does, I feel like I’m borrowing the experience and expertise of YC founders to build better software. How do the Matt Pocock skills compare to gstack? I am interested in any tools that help turn vibe coding into real engineering at higher levels of abstraction.
>>109234218i am coming in from the perspective that until electricians that start out making 70k (yes you heard me right) in this market; that until theyre not extinct from the common market; then all is good.
>>109234224you are retarding
>>109234252Ok, but I'm not an electrician. Even if it manages to replace half of all knowledge work (i.e. work that is done clicking and typing on a computer) that's still huge.At this point I would hate to have to get an on-site job.I only care about AI so I want AI related work or a bullshit job that allows me to spend 2/3 of my time on my own projects.
>>109234298What sort of projects are you interested in finishing anon?
>>109234310Yes
>>109234234What's so much better about their software?Reading the list of their success stories it all seems the kind of stuff that could be replicated in a week or two except you would have no users and they only succeeded over their competitors from shilling their own saas harder than them until the network effects kicked in.
>>109234234i just use /grillling to replace plan mode basically
>>109234336Can you not use slurs when talking to me?
>>109234342No because even if I apologize you'll just continue holding grudges and not cooperate. Thats the definition of a woman.
zucc is back?
>>109234310I'm making an LLM inference/finetuning engine, doing finetuning experiments and making synthetic datasets.This is one of the things I'm working on right now. I explained it here >>109232870
>>109234364>Elon and Zuck had enough time and compute to distill Fable, and now they're rolling out their models
>>109234364>deepswe 43 point jump
>>109234364Everybody's got a SOTA model these days...
>>109234328I’m not going to defend YC backed slop startups…But I like that gstack puts emphasis on defining scope/stakeholders/deliverables and helps you make informed engineering decisions before letting the implementation agent run wild. Vibe coding is only valuable in the real world if you have confidence in the output.
I'm making a roguelike. At first I thought I was doing something dumb because I was telling the LLM to implement the same things I've implemented a hundred times in my abandoned projects.But I soon realized I had zoomed past all that boring boilerplate, the code is architectured as I need it for my future plans, and I'm having a blast implementing features and playing with it.I'll probably not publish the game (or hide I used AI because of the stigma), but damn this is fun as fuck. Games are one of the most time-consuming programming projects there are, and I've been waiting decades for this (I actually have a 10+ year design document for this game that had been sitting inside a folder from my college days I'm finally glad to implement).
>>109234434To be fair Opus hasn't been SOTA for a while now
>>109234454Release it but don't tell a soul you used AI. It sucks but you're just covering your ass.
>>109234454Which game engine do you use, anon?
>>109234483
>>109234454Yeah AI will let us make all the stuff we thought we'd never have the time to do.
>>109234490fr when I started doing development work as a job, my personal projects became a slog and I hated my life for it
>>109234454>>109234472>>109234490No one is going to know or care if your code is developed with AI. But the blue hairs will get testy if you use AI to generate the art, visuals, and sound.
>>109234296Why?
>>109234454>or hide I used AI because of the stigmaYou should not be embarrassed about things like that imo. Everyone will eventually accept that AI is capable of building software, some just live in denial for longer before that happens.
What happens when you run out of stuff you thought you never had time to do?
>>109234468it is still a VERY GOOD model that can be confidently used, and with a proper harness you get flawless execution.
gpt 5.6 when? still not in my codex
GPT 5.5 is being unusually helpful with something questionable tonight
>>109234554I've seen the "he probably didn't even read the code" comment many times when people find out you used an LLM when writing code. Professional software engineers know this is as stupid as getting your panties in a bunch because someone used an LSP or auto-complete, but still. Indian-authored slop PRs are a real problem, so it's not like I don't understand where this comes from.
>>109232870>>109234365That looks promising anon. Sorry for lashing out at you I deleted my comment with the slur. I wish I could dialogue further and prod at your process a little more but I've pulled an all nighter so I'm gassed. Maybe you could anticipate my nerdy questions and I'll read your response when I wake up.
>>109234570Agreed. You can use it responsibly and still show craft by doing proper review (and documentation if it connects to other services). In two years no one will care if you used AI. In fact they might find it weird and "unsafe" for not using it in some manner.
>>109234554I made an update video where I thank ChatGPT for doing all the stuff for me and basically only this tranny seethed about it.
>>109234580At that point you realize you've been dead for 200 years and have been reliving your life in purgatory and decide whether to reincarnate or create your own harem planet reality with your overdeveloped programmer psyche. Because I can't imagine running out of wild ideas before I die.
Does 5.5 use caveman mode internally?Sometimes it glitches and it seems its thoughts leak, it's not the first time it happened, and it looks like this
>>109234615have been for over a year. not a secret.https://github.com/aw31/openai-imo-2025-proofs/
>>109234598I'm not the guy you responded to in the deleted post.
>>109234580You probably move on to overly ambitious ideas you always had like creating your dream GTA+Sims+SimCity+Battlefield+Europa Universalis game.
When working with codex specifically: stay on one chat session trying to leverage cache hits + accumulated context knowledge about the code base, or start a new session for every unrelated feature?
>>109234594Probably 10 pacific time.
>>109234595Update: it fucking delivered. Now I just need Deepseek to finish this off.
come onnn sol i need you don't make me use fable for this I don't trust it won't reroute me to opus
>>109234679details you filthy pig
>>109234683dont steal my crypto vibechud!!
>>109234683GPT 5.5 on low is all you need, really.
>>109234703I’m not stealing anything I just want to sell lewd games without bowing down to mastercard
>>109234683Lol howd you guys get access to sol early
>>109234703>>109234683since when did /vcg/ get into crypto? and by "sol" do you mean solana? I thought sol was some new model I missed over the last few weeks in my absence.
>>109234713I did not, that’s the problem. 10 pacific is the livestream so I’m waiting like the rest of the plebs
>>109234717I don't think /vcg/ got into crypto, I am building a crypto backed platform but that's just me. And by Sol I meant GPT 5.6 Sol in that post
>>109234722im sleep deprived and didnt read your image. anyways since bitcoin and cardano are cooked; why do you guys recommend solana again?
>>109234732Solana has lightning fast transfers and basically no fees. ETH L2 is amazing too. The whole architecture is pretty much there and ready to use, it's just that almost nobody is using it.One big problem is that we have to deal with the baggage of cryptobros hyping NFTs and teenage traders trying to become the wolf of Wall Street by flipping memecoins
>>109234698IPTV DRM bullshit
>>109234745How’s that questionable? Are you cracking a TV feed?
>>109234604Vibe coding (as initially defined, like not reading the code at all) will eventually become the norm as models improve. I already find myself reviewing the LLM code and finding no issues at all 99% of the time at work. Guiding it with architectural guidelines is still very important, and the main difference between a maintainable project and one that collapses under its own slop a week in, but that's just because we are the ones with the ideas and long-term vision for the project, and the LLM is the code monkey.As someone who started reading programming books when I was 4 (yeah, sorry if you find that hard to believe), this makes me uncomfortable because I can't imagine getting to where I am now if an LLM wrote all the code for me all my life, but I'm sure life will find a way and developing software will become a way to bring ideas to silicon rather than trying to translate them to programming languages "by hand".
>>109234745Oh. For watching illegal porn, I hope.
>>109234750Pretty much. LLMs get pissy when you deal with DRM.>>109234762Nothing that fun. The big boss man is complaining CNN isn't on his TV so I need to have this shit working again by the morning.
>>109234786>The big boss manYou work at an interesting company.
Should I use Terra as my new reviewer agent? I'm even considering keeping 5.5xhigh for that, since it works so well. I always use several sub agents, so maybe Sol is too much?
>>109234786>Nothing that fun.Pedos deserve the rope
>>109234814Terra is supposed to be the same but twice as cheap, so why not.
>>109234844Porno involving children isn't the only illegal kind, bro. Where I live it's all illegal!
my gpt 5.6 floppy just arrived
>>109234849Korea?
>>109234844You just outed yourself as obese (north-American).
>>109234871Nah Philippines
>>109234876speaking of obesity does someone know of a jailbreak prompt to research peptides. i got a sprained ankle and i need to study and test bpc-157 and then have agents scout the web to find me sources i will then thest.
>>109234923Are you sure you need a jailbreak for that outside of fable?
>>109234923protip: solve problems by avoiding them in the first place, not looking for solutions after the fact. this applies to every layer of the simulation
>>109232574its nuts how much more performance you get out of c++ instead of python
>>109234961with python you will bump your head a million times trying to get some library to work like how you want it to vs with c++ you just describe what you need and it makes it from scratch and it works insanely better
>>109234961careful this is a trigger topic here
>>109234961What I wanna know is why do some freaks still act like LLMs can't handle C++? It works fine, I think they are just coping for reasons not to use it
>>109234974I program a lot in C++ and I still wouldn't use it for most projects. Cmake is horrible even for LLMs. But I do agree that in general it becomes less and less of an issue as AIs get stronger. The problem is, say you want to code a web scraper. I can get one working in 15 minutes in python with an LLM. With C++ and the same LLM it would take me from 45 minutes to an hour for absolutely no benefit whatsoever.
>>109234756If you care for and for whatever reason want to display your craft in software development, you can do that even if using AI.You're correct. The area of expertise and industry will adapt to the new paradigms. I believe we will keep abstracting away from the code. Maybe you will focus a lot more in engineering and architecture, which also can be displayed but now by proper documentation, personal writing of the capabilities of the software, limitations -- anything that shows you understand what is going on and the primitives behind it.
>>109235000>The problem is, say you want to code a web scraper. I can get one working in 15 minutes in python with an LLM.Obviously if you are making jeetshit you should use a jeetshit language, duh
>>109235000because you're just reinventing the wheel
>>109234966>it makes it from scratchYeah but then you have a blob of a thousand+ lines to do that that you wouldn't have with Python. Fine if you're never gonna look at the code, otherwise it's rather inconvenient.
>>109235000my experience has been that its good to prototype in python but then once you know what you want a c++ version is better in most non-backend cases. python is good for like a web server that creates data that a c++ program uses. i built a simple program that counted +1 and plotted the count, python did 200k/sec and c++ did 250 million/sec
soon
chatgpt and codex will merge somehow todayso new voice chat can use codex now
>>109235043> i built a simple program that counted +1 and plotted the count, python did 200k/sec and c++ did 250 million/secusecase?i use python for practical thingsif I need to log and event or update an internal db why would i care about millions of writes?it's just retarded, it's measuring numbers for the sake of measuring them, a measurement in the vacuum is practically uselessi do admit all these language wars become more and more irrelevant as ais get better and I think we're way past the point where a language is anything but a personal choice
Chat im sorry i couldnt make it to the reveal...all nighter...too tired...my tokens never ran out....take over for me anons. remember me when you make it.
>>109235066im building a personal tradingview alternative that isn't limited with their 40s timeout cloud compute bs, and instead can do machine learning or other heavier scripts that otherwise wouldn't be possible. and also a few others basically in the same arena, a few cmd window python only terminal style. python works well with yfinance to build the datasets, but when trying to use various lightweight charts/highcharts/etc libraries i find its a lot easier to just prompt what you want with c++. it still requires a lot of patient bug fixing of course, but im not an absolute beginner so i have done it before.
Where 5.6
OP should be updated. Gemini CLI is already end-of-life.
>>109234967kek
>>109235066Yeah languages are so irrelevant that now I ask AI to make scripts in Python just so that I get used to Python syntax, not because I like it or think it's a good choice.
>>109235168Soon? https://openai.com/live/
>>109235189What's their new thing? That's only for $100/mo subs? "Sparkle" or something?
>>109235066Any language will get the job done, but there are still better and worse ones.I don't see any reason to handicap yourself, when AI is getting better you should create more complex projects instead of using suboptimal tools.
>>109235234measure worse into an actionable metric first and then we'll talk
Over the next 24 hours?! You said thursday! Not "thursday or maybe friday".
THIS GUY SUCKS AT TALKING
aaaand muted>>109235287it can always get worse
>>109235287who cares at least he doesn't have an Indian accent.
>>109235221Who are those nobodies and why are they yapping so much? Just give me 5.6, noone cares about pretty pictures in the web app.
>>109235287He's fine.
She should click on Finish update in Chrome. Security risk. ffs woman.
>>109232574>tfw company announced full RTOI have 2 months to make at least 150k ARR with my vibecoded app or I have to go back to the wage cage
tibo my nigga get some sleep
That's Tibo? He didn't look that chubby on Twitter.
what if they're all AIs?
chatgpt work? just call it workgpt
they better have a way to go back to the codex branding or i'm switching to claude.
5.5 found a hidden dev setting in my Xiaomi air filtration units to reset the filter back to 100% lmaooo
say goodbye to 5.4 anons
>>109232666I read this and I dont see how frontier models being gated by the government will continue to be a thing. Its just that the current admin is retarded
>>109235350damn, infinite free filters
>>109235355are they going to jew us with another 258k gimped versoin of the model?
>>109235350how do u even connect an air filter. thata ctually reminds me i can just fucking root and hack android linux shit basically with these models easily at this point.does anyone have a usecase for a spare android moto g play 2025? hard mode dont say server.
exoplanets anon they stole your idea
>>109235371make it into a remote emulated mouse+keyboard
>>109235372kek
>>109235362You bet.>>109235371They're smart air filters. You extract the token and then you can connect, poll, probe and send cmds to them from any host.
>>109235382that sounds retarded. so what i just pretend the phone is a mouse while i game csgo? get real.
>>109235350what's the point of an air filtration unit with a dirty filter?
is Work the app for codelet fags?
wait what the fuck
>>109235395I mainly use them for dust filtering so they easily last years with a bit of cleaning every couple of weeks. Not living in a heavily polluted area like Dubai lol
>>109235350are you being frfr
>>109235391Before I got fired my plan was using that plus an HDMI capture device to use external AI to automate my wfh job without IT bitching at me for installing unapproved software.
why don't I have a researcher wife like that
oof
>>109235371>does anyone have a usecase for a spare android moto g play 2025?wondering the samegot a new phone to replace my motorola g82was gonna put hermes on it, but i already have hermes on a vm on my server and i already dont use that so i don't think i need anotherwifi camera? the camera is better than the shitty 720p wifi camera i found in a draw and got working a few days ago, but i don't have much use for that either. I just watch recordings of myself at the pc.I dont know
>>109235420wasn't 5.6 cheating a lot though?
>>109235408Absolutely. There's a lot of stuff in there, can even read the RPM of the motor / volume filtered etc. which isn't even reported in the Xiaomi app.
Why is the camera swinging back and forth it makes them look like someone is on a rocking chair this is so annoying why are they doing this frucking cameraman
she is so cute bros, I bet she's multi millionaire too..>>109235428that's just one benchmark
>>1092354205.6 apparently was apparently cheating so i need to see the other benchmarks and also use the fucking thing before i make any judgements. sincerely hoping it's better than fable though because i can't fucking afford fable at API pricing
>>109235395it blows air >>109235409ok epic >>109235427ive been using vdo ninja (two words look it up) It turns your phone into a camera and/or a microphone. it works fantastic when you turn off all the mic customization and just leave the mic as clean as possible. my friends on discord say my voice sound crisp and perfect. Using phones for mics is underrated. the video works perfect too. its fully customizable
>>109235420>zucc and elon are now joining the circusthere’s no way anthropic will keep Fable locked behind API pricing, right?
>>109235468no cheating in deepswe
>>109235472>>109235427my bad i should clarify that vdoninja turns your phone into a camera/mic for desktop. or whatever you can manage to make it work with. i open a tab on my phone, start a session, put session link on desktop (for some reason edge works best for me mileage may vary) and then route it through virtual cable. adjust it on volume mixer. ez.
>>109235474I imagine Gemini 3.5 pro will be pretty decent when Google finally releases it as well. Competition is intense.
i don't care about the islamic republic of japan
kek
lmao nice voice demo
Did not realize they were doing this as I have only used Anthropic for the past weeks.We're gonna be eating GOOD bros
>vegetablesWhat?
He's a vibeGOD and he touches grass too.
did anyone get 5.6 yet
>>109235503that arrived to the cli very recently, like in last 7 days. used to be only the desktop app
>>109235519broken menu entry for terra
>>109235503No. They are jewing us. Why do you think 5.4 is going away specifically?
didn't notice they changed reasoning "low" to "light"
>>109235519yes
first the>fable is getting cut off guys! remember to FABLEMAXXnow the>the next openai model is coming guys! this is absolutely HYPEyou guys really hobbified this tool lolnot that I'm against it, you do you. I just see some unexpected similarities to other tech cults
>>109235556fuaaaarrk
>>109235550no one likes to talk with 5.4 desu too autistic
>>109235564i'd argue not autistic enough - 5.3 codex was far more autismo and i liked having it around.
it's over
>>109235559I do see that, but unlike the next orange iPhone this stuff is actually practical. If I could rent something like a 256 core CPU for 200 bucks a month, I would feel the same way.
No Sol for Yuropoors?
how much context can you get from sol on a free chatgpt account?
>>109235594I don’t know what kind of bench this is but anything that puts sonnet ahead of 5.5 smells like bullshit to me
>>109235598Based
>>109235564My point is they already took away the 1m context from 5.5, and now they are taking away the only model that exposed 1m through the sub.Codex compacting every 5 minutes is REALLY annoying after having used that one and Claude.Most people never even knew 5.4 had 1m ctx because you had to enable it through config.toml but also people who use Claude Code know how sooo much better 1m ctx is.
we're back
>>109235594GPT 5.6 so much better then GPT 5.5Feels good GPT bros
>>109235599I don't think you get Sol on a free account?
More info https://openai.com/index/gpt-5-6/
>>109235598>>109235602If that's true I'm signing up from a VPN fuck this shit.
>>109235624I'm in Europe and I have Sol now.
>>109235619i have no idea. i'm used to using gpt 5.5 on a free account. i get very little context and it blows through the weekly limit. but for quick debugging sessions it works. maybe i should just check myself.
>>109235624It's not true I am europoor and gave all my euros for the 5x Pro plan and got it right away.
>>109235594uh Terrabros wtf
>>109235601sonnet 5 is unironically better than gpt5.5, yes. gpt5.5 just isn't as good as you think it is.you're insane if you expected 5.6 to be anywhere close to fable though. like at least it's affordable? once fable is off subscriptions i'm moving to 5.6 sol, but yeah 5.5 just isn't good. sorry you were gaslit by brand loyalists instead of just using the best tools.
>>109235603Trvke, fuck copex sissies
>>109235594mogged by fable again award
>>109235598they're rolling out over 24 hours>>109235660>mogged by the thing that you can barely use because they make a worse model take over half of the time and still plan on making API only>mogged by something that takes 5.4x as longIt annoys me that they don't even say which effort level they used
>>109235655Gpt 5.5 is better than Opus 4.8 and few people have brand loyalty to OpenAI. Claude was the best model for a while and almost every serious vibe coder was using it, it's actually people who tried several models who switched to GPT when it overtook Opus.
Chat: Plus, Pro, Business, and Enterprise users access GPT‑5.6 Sol through medium and higher effort settingsChatGPT Work and Codex: Free and Go users access GPT‑5.6 Terra. Plus, Pro, Business, and Enterprise users can choose among GPT‑5.6 Sol, Terra, and Luna and set an effort level for eachmax is available to all users with access to GPT‑5.6 in ChatGPT Work and Codex and can be toggled on in settings. In ChatGPT Work, ultra is available to Pro and Enterprise users. In Codex, it is available to Plus and higher plans.
>>109235655>>109235657anthropic brainrot
Not having 1m context is disappointing.
>>109235685What's the difference between business and enterprise
is Codex plus even worth it?
it's no big deali can wait till tomorrow for sol hon hon hon
>>109235703You mean the $20/month plan as opposed to nothing? Yeah, I didn't max out on my Plus plan last week, and if I did I have 3 resets in store anyway. I'd pay for Pro if I didn't have those resets, but even with Plus I barely even need them.
>>109235619>>109235637thanks gaben
Claude just reset usage lmaoo
>>109235703yes and you get image gen and new voice mode as a bonus
>>109235721only have terra and luna availablesol is for payingchadsit's over
>>109235724kek based free market capitalism
Fucking claude reset finally arrived
>>109235724>it's realthank you dario
>>109235724>it's trueFINALLY THESE FUCKING FAGGOTS holy shit i can work again
>>109235657>>109235692Called itFucking jews
>>109235728terra only, I thinkluna and sol for plus and above
Lol at all those who bought yet another extra max plans until this claude reset
>>109235745no fucking way man LOL i'm actually so sick of this. maybe i'm not switching after all, working without 1m context is just exhausting
>>109235746Luna is the weakest model so I don’t think so
anthropic have their little claudekeks wrapped around their finger dance goyim puppets DANCE
>>109235760As much as I like GPT, this is true.
>>109235745uhhhh solbros....?
Chat is this true? https://x.com/i/status/2073460581001220370
I pay all the money where Sol? Sam where Sol Sam saaaar
I wonder if Fable will even really go away, looks like they can afford the tokens after all.
>>109235763i'm a localfag but i only have 128GB RAM DDR5 so no opus-like model for me yet. i need these labs to work faster and release weights for free.meanwhile i have to suck dario's cock. i don't like it but it is what it is
>>109235724Panic move
>>109235770save the bait to post in 1 month when everybody's bored
my weekly limits just reset when they reset yesterday???
AmeriCHADS how is sol? Is it all it was hyped to be? Tell us plebeians about it
>>109235764i'm genuinely questioning whether or not people who defend small context windows actually use the products they push. like what are they even using this for? frontend web dev? how do people tolerate this? is it just because they've never experienced 1m context? i would've switched to sol if not for the tiny context window. i just don't know what they were thinking, man
>>109234650new session for unrelated features
>>109235782dario won
>>109235784Can't, Fable rates reset. I'm busy now.
>>109235784I'm a Europoor and have Sol. I think it's just Pro subs and Plus get it 24h later
>>109235785sub-1M context windows are fine if your compaction is good enought. heard GPT 5.5 is fine for many long-ish-running tasks
IT'S OVER
>>109235785i think this is only defensible from a very "human-in-the-loop" perspective, where you're looking at some kind of jira agile board and solving tickets/bugs/features one by one and you open a new session for every new thing.i don't see how this little context can be properly used for any macro strategic planning, long brainstorming sessions and prototype building on the fly. 1M context is the king
>>109235776BAD GOYIM BAD BAD GOYIMwat are you doing here? your usage just got reset. You are supposed to be using fable right now. DO *whip* AS *whip* YOUR *whip* MASTER *whip* SAYS *whip* WHEN *whip* I *whip* SAAAAAAY *WHIIIIIIIIPPPP*
WE'RE BACK
>>109235808you can tell the AI to inspect the thread to soak up past decisions
>>109235745lol1m is half the reason I only use opus for dev work
Sol autonomously post-trained Luna
>>109235770demonstrably falsethere’s a guy here who, at least when he started, only knew “Git is a place where software gets downloaded”I think he might be exoplanetanonthat said, vibe coding works way, way, WAY better if you have a lick of software engineering sense (which is now getting decoupled from being able to program)- what Git is- what languages are good or at least acceptable for a given tasketc.
>>109235835>there’s a guy here who, at least when he started, only knew “Git is a place where software gets downloaded”thats mei still only vaguely know it has something to do with making revisions easyi just tell codex to undo changes and it does iti dont think i need git
>>109235808nta, butyou just described my workflow and I only use 1m window modelsevery time they reach over 30% I start a new session
>>109235852you might not need Git but it’s a useful tool for you and your clanker because your clanker can use it to stash (that’s the technical term) changes and go back to the last checkpoint and flip back and forth to diagnose fucky wuckies (not a technical term)and having Git history will let your LLM understand stuff like “over the past couple of days I did X, and now ____ needs even more updates” and it can go through the Git log and see dated commits to know what you’re talking about
j-just waitGPT 6 will totally beat and make fun of fable
>>109235856>every time they reach over 30% I start a new sessionwell so you effectively don't use 1M context.it means Sol was released with your use case in mind, apparently. less strategic and more executioner.
>>109235852>>109235835You absolutely do need git. I'm a no-coder too but with any work I do that involves creating or manipulating existing code, I always ensure get tracking is in place. It functions as both an edit history for the entire project I'm doing but also helps the model unfuck any mistakes it makes. Due to their very nature they will make mistakes here and there and make changes they should not have made. So git is used to help me tard rangle the model. This is especially important post compaction where it may forget subtle information about the project
>>109235815oh fuck oh fuck
>>109235877when my 1M context gets down to 70% I get twitchyI think the advantage of a 1M context is that you don’t have the risk of an intelligence drop-off right after a poorly-done compactionand a lot of my more important tasks take more than 20% or 25% context of a 1M window
>>109235877>well so you effectively don't use 1M context.I domaybe this is stale knowledge/experience because I started way back in '22, but the higher percentage of your window gets filled up, the more likely mistakes and hallucinations becomea 1m model at 300k tokens is less likely to get retarded compared to a 300k model at 200k tokensalso I avoid compaction like the plagueagain, my knowledge might be outdated, but it's worked pretty good so far
why did my CC weekly limit just reset?
>>109235796it's not terrible but I have a workflow that asks the model to read a large set of .md docs on each compaction so even with Sol's new slightly larger context it still occupies 30% of it right away after compaction (used to be 40%)and sometimes there are time sensitive things>>109235785Most codex users probably never used 1M window. But in the case of people who really should know better it's really baffling.It's almost as if everybody forgot we were supposed to get 1M gpt 5.5 context "later on".
>>109235802Mathbros... not like this...
>>109235912That’s still true, but I think unlike codex it at least won’t auto-compact your seesion which makes it worse.
>>109235934>everybody forgot we were supposed to get 1M gpt 5.5 context "later on"I have no issue with compaction 9/10 times, xitters also talk positively about it too
>>109235874What if Fable 6 comes out before GPT 6?
>>109235929TURN ON CNN
>>109235802>>109235935sorry nerdsthe x-risk of math is too highit had to be nerfed
>>109235948xitter is retarded
New thread, mad lads:>>109235994>>109235994>>109235994
New Codex app UI is looking great
>>109235934>Most codex users probably never used 1M windowBut it was capped to 272k for 5.5 anyway, was it not?https://github.com/openai/codex/blob/d72d669ca727a26765315ffe03db3dd2594d9b91/codex-rs/models-manager/models.json#L359-L360And 5.6 is capped to 372k.
Sol Ultra chugs tokens like there's no tomorrow, Jesus. Many times what 5.5 xhigh did.
>>109236001Yes, that's what I've been bitching about this whole thread
>>109235785Since Fable I was using Codex mostly for reviews, with sub agents it fills its context once for a pretty big review.Before that it was just a tradeoff. Opus has 1m but I still think it's worse than 5.5, it just makes too many mistakes, even if GPT compacts once or twice the results in my project are still better.
Gemini is pathetic. Why the fuck did Google focus on speed?
>>109236001>>109236020you could just uncap it if you wanted lol
>>109236223not how it works5.5 at least was capped on the server endpointonly 5.4 could be uncapped
