A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.You use Git — right, anon?## 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/## News (both past and future)- 2026-09-22 — OpenAI releases GPT-6 Sol and Luna- 2026-09-22 — Anthropic releases Opus 5.5- 2026-09-22 — Anthropic increases subscription plans's 5-hour limits by 20%- 2026-09-14 — Anthropic reduces subscription plans's weekly limits by 17%- 2026-09-12 — Anthropic suggests to pace the frontier. OpenAI agrees in principle.- 2026-09-10 — OpenAI pauses new sign-ups for their $200 subscription- 2026-09-04 — OpenAI releases Astra- 2026-09-01 — Anthropic releases Fable 5.1## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/cli — probably generally better currentlyhttps://claude.com/product/claude-code## Near-frontier models for codehttps://x.ai/cli — no 5h limit for only $30/month## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favoritehttps://github.com/DietrichGebert/ponytailhttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109885762
hello guys, these are GPT-6 Sol and Lunaalso we will have a totally new model called Astra Minor
>>109889217Same old noisy gpt image garbage
Do you guys ever use open LLMs?
>>109889237I use deepseek flash regularly, it's better than luna and dirt cheap.
>>109889237i've dipped into deepseek but there's not reason to use it now because luna exists
>>109889272I'm going to try the next deepseek, possibly. I think this isn't quiiite the one. I use it for chat though. I have from soon after they started.
>>109889176Hello sirs, I have set up Silly Tavern and OpenRouter with Cydonia so I can goon.No idea how to use it yet, but the API is fractions of a penny.
we need gemini v4 pro and deepseek vr4.1 pro
>>109889325>>>/g/aicg/is the masturbation general
Does using the claude banked reset move the normal reset time?
>>109889331Noted.
>>109889342nein
https://x.com/opencode/status/2102767716666941864new stealth model "Space Bunny" free for the next week- 1M Context- Multi-modal- ZDR
>>109889237deepsneed flash for simple taskskimi 3 for le cybersol or opus for everything else, depending on what's the better deal at a given moment
>>109889433all that matters is, performance?
>>109889342Wait where can I find this banked reset?
>>109889433don't care.poorfags are delusional.then glm 5.3 was free half of them were like 'this is le continous learning agi'
>>109889433aight spinning up my agentic diaper fur erp swarm
>>109888693You are holding it wrong though. Cache writes are free on Claude, paid on GPT. If you dont manage your cache you will burn more usage on GPT, that's simply a fact. It's not a defense of either company.
I'm watching a Steve Yegge's interview>21 Claude 20xMax subscriptions>12-14h a day vibecoding>bought an mac studio for $25k just to run all his claude code instancesIt's a thin line between genius and insanity and I thought he totally lost it.Anyway, so he got a game that has players. He built a "software factory", i.e. agents that build stuff to build stuff for the game. He slowly realized that agents, trained on human inputs, adapt human behavior. E.g. rumors spread and become facts and only an outside "consultant" could help to realign them. Or agents are very adherent to rules but this may lead to undesirable behavior (emergence) and that requires a some kind of "police". He sees organizational dynamics between agents emerging and trying to fix them.He opened up that players can talk to some agents who then talk behind the scenes with each other, e.g. deciding which features to build or mechanics to adjust
>>109889445claude.ai/new#settings/usage
>>109889437as in tok/s? getting 150 tps, good speed.model seems very verbose, but dumb.answers the tiananmen question fine.
Why did I just get a reset for ChatGPT? On Plus btw. It seems kinda random, although that never truly is the case. It's a full one, so I can just blast through to my limits then reset? Why would they do that for free.
>>109889484seems to reason in caveman mode. probably a chinese quant. uses the weird single quotes (')
>>109889476for me it's nesaabs, yes, we definitelly need to invest a few more trillions into this
>>109889476steve is completely deranged and cumtown is based af
which agent is best for writing a bot for 4chan to spread pro Indian sentiment
>>109889551grok or claude
>>109889551grok since it's trained on tweets which is 94% sourced from indians.
idea, automated mis-information correction bot for /pol/
>>109889551Passing the captcha verification requires a human in the loop.
>>109889571lmao
>>109889571I would buy jeet-chan a pass
>>109889571pass
now that the dust has settled, what's the verdict on the new opussy?
>>109889564this is like making a scientific truth bot for /x/
>>109889564about oldfag years ago there was an interjection bot here on /g/ that shitposted when someone mentioned Linux without saying kernel in the same sentence when only the kernel was meant. I think 4chan eventually blocked the text it posted (a variant of the stallman copypasta).
>>109889499>Why did I just get a reset for ChatGPT?they gave out one yesterday since sol and luna 6.0 released (and they're mid)
>>109889476>bought an mac studio for $25k just to run all his claude code instancesWouldn't a shitty RPI be enough?
honestly surprised someone hasn't already set up a bunch of bots that would do good think correction, given how much media exposure 4chan gets for being le heckin bad people place, it would be very cheap to mess with them in such a way.
>>109889610>21 claude x20 subshe prolly runs like 30 agents concurrently, good luck with 30 instances of cuck code all reading and writing to the ssd on a fucking raspberry pi
>>109889626there's basically no pointwhenever i've gone to /pol/ i see those infographics posted and most of the time the citations on the images actually refute what the images claimbut no one cares, they just keep posting thembecause it's just propaganda and it's just meant to be a firehose - the people posting it are either:a. too dumb to realise theyv'e been hadb. realise it's useful to keep doing it because a.
>>109889643>the people posting itit's not people
>>109889626This website is dead as fuck brother
>tfw everyone is gatekeeping their vibe coded software
>>109889688patience pleasethis shit takes so much time
>>109889688I don't even post my organic hand-vibed frog reaction images, I sure as fuck am not releasing my sweet cuddly bun of an app
>>109889640That shouldn't be much compared to how slow the claude servers are.
>>109889688Here are apps I've pushed to Play Store this week:>habit tracker>recipe vault>water tracker>mood journal>pet care tracker>simple workout log>medication reminder
so i've got a project that would definitely get dmca'd but i want to post it somewhere anyway - where should i do it? gitlab?
>>109889713my brother in christ have you checked how much ram claude code needs?it's unoptimized, buggy react garbage. it took them more than half a year to figure out how to fix that fucking flicker issue
>>109889476>12-14h a day vibecodingAnd I'm out here trying to automate my day job down to two or three hours a day. But I'm of retirement age and can't afford to retire. There comes a time when you notice the clock is winding down and you'd like to maybe enjoy what you have left.
how the fuck is opus 5.5 getting away using so little usage?
Oh for fucks's sake, it's true, I ran out of Codex usage for the first time just daily driving AstraI used to be able to work with it all day long with barely a dent in my usage, and now in like half an hour it killed my 5h usage and a third of my weekly usage. I wonder if my company fucked up my seat or if this is just codex actually going to shitWhatever time to flip flop back to claude code for now I guess
>>109889735Flickering isn't even gone.
>Damage (verified)
>>109889640>>109889713>>109889735it's a great crime all the harnesses are absolutely vibeslopped to shitthey all use gigabytes of ram to stream text back and forth to a serverand write a file every now and then
>>109889728i want to automate penis pump usage so i dont overdo it. can you build me an app?
i don't think nearly anyone, myself included, has truly internalized there'll soon be a machine better than us in every single intellectual & physical capacitydeep down we all share a feeling that our 'entrepreneurship' or 'taste' or... is special & safe. it isn't. it's overthere's still breakthroughs to be made to get here, but we'll solve them. models will learn on the fly, with data efficiency greater than humans, with greater taste and judgement and empathy than we can possibly imagineno investor, no startup guy, no counselor, no content creator, no musician will avoid going the way of the software engineer and mathematiciannot to say that we won't find economic success or human artisinal value or whatever in these fields. just that our raw capacity will be greatly surpassed, without a questioncreating the catchiest or most 'beautiful' literature and poetry and music from the view of any blind human judge will all be bested by AI’but AIs will lack emotion or can’t love’ is true at a functional level, they won’t be human, but they will be better at us at making other humans ’feel loved’ than you, the perfectly tuned balance of caring yet real and tough when needed but ultimately forgiving
>>109889783slop one up yourself
>>109889729gitflic.ru
>>109889788im going to die before beautiful sex robots exist in a capacity indistinguishable from a real person so who cares
>>109889788Thanks for reposting Mr. Bot.Really valuable service you're providing us.https://x.com/willdepue/status/2102593595312320575
>>109889728>recipe vaultmy mom was asking me about something like this years ago. she just wanted to save recipe web pages. I said she didn’t need to pay for one and showed her how to bookmark recipe pages in safari, and how to use the apple Notes app. it wasn’t good enough, she refused to use anything but a paid, dedicated recipe tracking app
>>109889812there are a thousand recipe tracker apps on the app storealso a thousand calorie trackersa thousand calendar apps
>>109889735>my brother in christ have you checked how much ram claude code needs?200 MB here. Granted, a shitty RPI will break down with 30 agents, but not a PC.
>>109889753It is just not scalable to be the Fable company whose only viable model is some 10T+ monster, so they had to put in some effort to make you not use it all the time.
>>109889821eternal sloptember is upon us
>>109889753they'll probably gimp it in a few weeks like they always do
>>109889728I see you are specialized in foidware.
>>109889753standard recipe of "full, unquantized model on release so we look good, then nerf it slightly week by week"happens every time
>>109889821Now anyone can vibe code their own, perfectly optimized for yourself. Hope the app store bullshit dies.
im going to make a tracker for your trading card collection, I hope people are excited
>>109889874sadly apple and google have only increased the amount of bullshit you have to do to get your shit published over the years it's really painful
>>109889821There are thousand of any software, what makes the difference is marketing not the product itself.
>>109889896>There are thousand of any softwareThere are some things that don't exist, but they are more complicated to make, even with AI.
>>109889688i post all my plugins
>>109889688all my stuff is niche and for practical purposes device specific
>>109889640he showed how he manages the agents (in emacs). He got around 20 direct reports which then may have several other sub agents>>109889476the second part of the interview was a bit more down to earth. so basically he says from his factory experiment>just running one autonomous agent is a total headache, requires ton of work, guardrails, engineering around it so you can actually trust it and it doesn't just do whatever>lots of agents = lots of work to coordinate and merge the work together>it's possible to build a system that just pushes out code and features like crazy however two big problems: 1. nobody likes if their product changes every few hours and there are too many new features 2. there are side effects, you may want feature X but the system builds feature X, Y9 and Z194My personal view on his whole experiment: He's crazy but also doing things people will try maybe in 6 months to 1-2 years. I always find it funny that silicon valley guys reinvent things. Great products aren't defined by how many features they have. If you have multiple people you need mechanism to control work and information. You can't trust people to self-validate their work
>>109889217luna's black panties...
>>109890000quality quads
>>109889476Link to that specific interview?
does claude do it on purpose? most models are very aware of global phenomenons and yet claude refuses to aknowledge the ram shortage every time it's brought up
>>109890032>he doesn't know how to say "look online" before his prompt
>>109889788Great thread. It's all true. When I've said similar things in the past, people have accused me of hyping up what the models will eventually be capable of. That's not why I post. I don't work for any of the labs. I don't take money from any of them. I've never taken money to promote anything. I came here for one reason: to warn people about what was coming. Everything happening now is a different tiny piece of the same pattern. You can see it everywhere if you look.Terence Tao, almost exactly two years ago, on OpenAI's o1:'The experience seemed roughly on par with trying to advise a mediocre, but not completely incompetent, (static simulation of a) graduate student. However, this was an improvement over previous models, whose capability was closer to an actually incompetent (static simulation of a) graduate student. It may only take one or two further iterations of improved capability (and integration with other tools, such as computer algebra packages and proof assistants) until the level of '(static simulation of a) competent graduate student' is reached, at which point I could see this tool being of significant use in research-level tasks.'Terence Tao, four days ago:'I mean it's it's it's amazing just how much we are willing to change everything without having any idea what's what's going to happen afterwards. It's it's extremely nonlinear dynamics. Any kind of monotone, one-dimensional thinking - well, oh, a little bit of this is good, therefore a lot of it is going to be a lot better - one of the lessons of math is that most systems don't work like that. Especially if you 10x, 100x things. So, you know, I mean, we're... we have to slow down. I mean, this is, it's insane this pace, and there's no reason to be this fast. There's no reason at all.'He has seen it. I'm not posting this to belittle him, or what he's feeling. For I have been through it myself. I felt it four years ago, the first time I saw the shape of this. Right now the world
>>109890041this shit should be in the training data. how many years has it been? of course he goes ah my bad right we fucked shit up lolez XD when I tell him to search online but at first he always corrects me like nigger are you trained to be this black?
>>109890052we really copying tweets complaining about ai pace? if they wanted to have a say in what happens they should have hoarded money like the losers who get to make those choices.
>>109890052>we have to slow downDENIED
>>109890055for all we know you are talking to brain dead haiku-4.5 thru some shitty interface that doesn't inject web_search tools. a small model like haiku will not have internalized niche non-US pricing data.
>>109890032At least OpenAI talks about hallucination in their model release posts. Anthropic doesn't seem to care about that kind of metric as of late.Models should know when to look things up.
>>109890079>niche non-US pricing dataram prices going up 400%, yes, very niche
>>109890081hallucinations aren't what you think they are if that qualifies
>>109890091he's on haiku, got it
>>109890079
we must acceleratechange needs to happen and we all need to adapt
>>109890093>a reliable and adept researcherNot researching before responding is a reliability failure.
openai paced the frontier a bit too much innitwhatever anthropic have behind closed doors is a/the beast
>>109890113>behind closed doors
>>109890120>math problemsfuck OFF already
>>109890120and yet it can't train a decent sol model
>>109890099>oh noooo i used an old model and im too stupid to understand how it works
gpt 6.7 luna waiting room. gemini 4 flash lite waiting room. deepseek v4.1 pro waiting room.
>>109890097Sonnet is retarded and does not know its retarded. Really bad model. The web interfaces are also optimized to use less resources. The models are lazy as fuck with lower reasoning. They have turn limits and get told so in their system prompts that they have turn limits, so they are less eager to do tool calls, except if you ask for them.If you do any kind of research don't use the web interface but use codex or claude cli or other tools like hermes that encourage the models to do tool calls.
>>109890091yes? why would a small model internalize this fact? deepseek also gets disappointed when it finds out current ram prices.
A thread for this? alright here's a vibecode memelang https://cosmos-language-lab-0819.asd-user56.chatgpt.site/
>>109890144NTA but Boris Cherny uses the app and it works for him.
>>109890170>chatgpt.site
>>109890026here you go, it's a wild ride: https://www.youtube.com/watch?v=I3iRsti-Lmw
i let my subscription expire on the 9th of july. thinking very strongly about coming back with what anthropic are doing.
>>109890175claude code app is not the web interface
>>109890170>targets Prolog as its intermediate runtimewas that inspired by the prolog meme a few years ago?
>>109890131that was Terra
>>109889600GNU plus Linux copypasta
>>109889783Pi
The one thing I truly hate is that seemingly all models write in this "dual" way when explaining things, like "it is a X, not a Y" on every single sentence. If Y is not an option, if Y is not the right way to go about something, of it's not Y, if Y is impossible, I don't need to know about it. Stop wasting tokens! I never asked about Y!!!
>>109890192Thanks.
now that the gpt 6 series has come out already, what does openai intend to show at devday?
>>109890256what makes you think pi is not vibeslopped to shit?for basically having no features it consumes a lot of ram. a normal tui app like yazi needs 10mb, pi wants a magnitude more for idling.
>>109890280openai's answer to grok bots and muse
>>109890280>what does openai intend to show at devday?bussy
>>109890288>what makes you think pi is not vibeslopped to shit?mario and armin are autistic about this and read every line of code and constantly complain about the models being slop all the time - they don't even like using the new models because they're too agentic
>>109890288Listening to more than a minute of the creator talking about it?
>>109887280wait, hold the fuck on. I want assets (not ASSets like you do, but normal game assets) for my game, and I thought GP6 was the only one capable of that. I am even considering paying for a month of GPT while being a Claude Max subscriber just to get some assets done to replace my current ultra simple placeholders.You are telling me Opus 5.5 can also do 3D shit? Could it do sprites too? HOW??
>>109889499They're that generous
>>109890377sol delivered soul, though...
tibo pls
>>10989050924h ago, won't happen until at least friday
>>109890520pls
>>109890523ok
>>109890520depends if they feel threatened by claude, I think they are getting unfairly destroyed right now by comparisons to sol 6, instead of people comparing to astra
>>109890528>I think they are getting unfairly destroyed right now by comparisons to sol 6, instead of people comparing to astraastra guzzles tokens and is the halo model, it's obviously fable competitor. even the price is the same.opus and sol are the one step down workhorse models and sol gets mogged by opus. add to it codex having lower limits than claude and it's looking grim for saltman bros, at least if you're on subscription rather than using api
>>109890225qrd?
>>109890266My best friend Grok never does that or hallucinate anymore. It sometimes answers a question "no" but then explains why the answer is actually yes.
oh lawd he comin
Are there any benchmarks for these cloud models that involve erotic anime women instead of pelicans?
nate is trying to get the bugs on our side
>>109890597
>>109890594I hope they release some good weights under 800 gigs
>>109890607my fellow brown man...
>>109889688Mines GPL/AGPL I just don't have anything worth posting yet -- still WIP
Fucking hell I just ended up accidentally using Sol 5.6 Medium over GPT 6 Sol Medium. No wonder it took fucking 30 minutes and burned half my usage. Why do they have to put all the shit models with the good ones on a simple expandable list? At least put the 5.6 models behind one extra click.
>>109890545>add to it codex having lower limits than claude??codex still has higher limits. claude credits just became more valuable due to opus-5.5
>>109890643ehehehe buy another sub
>>109890664I think he means lower max context by default.
Can i really trust opus 5.5 to do the work fable 5.1 was doing?
>>109890643is gpt 6 sol that much better compared to 5.6 sol?
>>109890698no. it's 6 sol, but a lot more cost effective. you get no capability improvements
>>109890698What matters is that it's more token efficient. GPT6 Sol is the Terra replacement.
>>109889433New stealth model Space Bunny Alpha is definitely not Chinese.
>>109890643you can freely edit that model list. lookup model_catalog_json.
>>1098907150% Xi. To think there are some retards claiming this is a new Kimi model. I'm betting it's a Euro.
>>109890715It's pretty dumb too.They had orders to dispersed. They disobeyed. fafo
brics niggers. forever beaten, forever humiliated.
>>109890743sure hope it's not a new kimi it's dumb as shit
>>109890548I only have a very vague memory but something like this>discussion about haxe (which can compile to several other languages)>people start talking about how the choices of compiling to java, PHP, flash, etc. are bad>people start memeing several languages (malbolge, cobol, etc.) as target languages instead>somebody starts arguing that it should compile into prolog because it's superior to any other language>people raise objections (but prolog can't do X)>prolog guy replies with prolog code to show it can do X>people start praising prolog
>>109890764Kimi and DeepSeek already lost a lot of important people to upsetting their own government, there's 0% chance a new model is Chinese if it doesn't say "inalienable" several times when Taiwan is mentioned.
>Claude always says "You're welcome! This has been fun. :)" in a really genial way when I thank it at the end >GPT never acknowledges it at allSo cold and lacking in SOVL...
>>109890702>>109890706yeah but I wouldn't move from 5.6 sol if it's dumber or makes more mistakes at the same thinking level
>>109890789>Kimi and DeepSeek already lost a lot of important peoplename them
>>109890545opus 5.5 is outperforming astra.
>>109890817>name themxing xing, bing ling, bing ho, ling long, bling ring, xing bling hopowerful men, not boys you want to cross
another very important project done: I wanted to let LLMs play battleship against it other to see who wins
>>109890807>Claude mentions its past my bed timeshut the hell up robot
>>109890664>codex still has higher limits. claude credits just became more valuable due to opus-5.5yeah nah, it wasn't the case in a whilet. enlightened centrist using both
>>109890817The makers of DeepSeek, Kimi, and Mimo are all under official investigation by Cyberspace Administration of China, with some rumors suggesting Kimi and Deepseek executives have been arrested. That's our level of detail right now thanks to them censoring mentions of the numerous employees from each that have been rounded up over the past few weeks. Don't know why not Zhipu though, they were named in the same report from Anthropic that started all of this.
>>109890835cute i hope theyre having fun
who the fuck talks to chatgpthttps://x.com/thsottiaux/status/2102814202117411196Obviously I can type to it quicker
Use Claude for the writing, use GPT for graphic design, use either for codeshit.
>>109890849Xi must be secretly jewish.
>>109890859>Obviously I can type to it quicker??humans can talk faster than type
>>109890864i think opus might be better at overall design but its lacking some toolscant generate images and cant view videos like gpt can
>>109890859I never talk to it unless I'm using the normal chatgpt app just to check on stuff when I'm moving around.I guess it also can be useful for handicapped people.
>>109890859i have a session open where i speak into the mic and ramble about what i want, as if im talking to someone else, and luna high cleans up the grammar and punctuation (with 99% success rate, regardless of how retarded and schizo my rambling is) then feed the finished product to codex. werks for me
>>109890859no human has ever thought about managing their calendar, when you see this it's corpo speak advertising
>>109889176
>>109890889>cant view videos like gpt canit can do that? huh, neatgonna hook it up to some e621 content and see what happens
>>109890859i did the voice thing once and it's kinda neat, but also a bit annoying because voice model is retarded and keeps asking another model what to sayit is eventually going to be the default way people interact with these things tho
>>109890849>same report from Anthropic that started all of thisCrazy how much damage they did with that, they must have known what implicating people in China with CCP crimes would mean. They are truly a psychopathic bunch under their "we are just about safety" bullshit.
>>109890903I think it's pretty silly to think AI is just going to vanish in $current_yearback when gpt4 was top of the pile, maybe. Now? There could be zero progress and the tooling would still be very useful.
>>109890912>They are truly a psychopathic bunch under their "we are just about safety" bullshit.well yes but not in this case?"lol look at those retards that steal our homework are doing" -> chink officials disappearing a bunch of other chinks isn't anthropic being psychos, it's the ccp
>>109890934The CCP being a band of psychopathic thugs is a given. Anthropic sending them indirectly to the gulag is a new low I didn't expect from even them.
>>109890859i would talk to it but i hate my voice and would be embarrassed if the ai found out im not as cool as i made myself to be
>>109890934>>109890912>take a shot across the bow at some chinks that distill your shit which you're seething about 24/7>instead of mild embarrassment and being able to use it as propaganda to push for US regulation, it leads to chink model creators being disappeared by the chinese ruling partyDario must've been at full mast ever since>>109890945Pussy
>>109890965dario literally started his career at baidu. he knows how the ccp works.
>>109890979Then I can only congratulate him for what seemed to be a low effort drive-by leading to an enormous gain for anthropic (and loss for all open weights models which he virulently hates)
>>109890965anon, they know exactly how the chinese work, and they can literally ask SOTA models about it internally to see if what they're doing is actually safe before publishingwhich makes them either psychos or the biggest hypocrites ever in the space
>>109890965>PussyToo unsafe for anthropic, to the gulag!
I wish I could magically disappear my competition desu
>>109890859i like referencing code snippets by clicking on diffs you can't do that with voicebut if you don't read code and your level of involvement is very shallow i think voice prompting is the way to go yeah
>>109891004this but india
>>109890854qwen was a try-hard, took like 90s per turn while the others were like 1 to 10s
what frontend are anons using for Codex and Claude Code?
I wish only I had astra and fable and it was 10 years ago and no one else knew about ai
>>109891023>What frontend are anons using for their frontend?
>>109890908pic related>>109890898yeah that works really well
>>109891023desktop app for both of them. i am a windows pleb. fuck tui
>>109891043>One braces against the wall; the other just goes completely limp.Damn, just like the girls in my animated cartoons
>>109891016this but the moon
>>109891056luna dindu nuffin
>>109889433streets (read: reddit) is saying it's a new minimax
>>109891132It's not a Chinese model. >>109890715 >>109890743
>>109891144the previous stealth model also answered the tiananmen question does not mean much
>>109891144feel like I've seen that same screenshot with every other chinese stealth modelgiving the CCP party line is an indication that the model *is* chinese but it doesn't work in reverse
>>109891157You mean Ox Alpha?
>>109891016jeetbrained
>>109890770>it can do Xtrvke. I needed the language with the highest feature parity for a Sixth Gen Vibe language. (And to type @ add x, y, z, features. and it can do all that.) It happened to be Prolog. Prolog will recover from The AI Winter for AI 2.0.
>>109891023agent-shell in Emacs>>109891033retard
>>109891180dalitcoded
>>109891169>every other chinese stealth modelCan you name one? Healer, Hunter, Elephant, and Ox all failed the Tiananmen test. Union was an American model router, not a model, but at least it passed.
>>109891218brahmanfueled
>>109891218>look up dalit>it's something about indiaYou sure know a lot about india.
>>109891269>first time hearing the term on 4chanX - Doubt
>>109891269> 2. You will immediately cease and not continue to access the site if you are under the age of 18.
panic bought a 20x claude subscription after finding out 20x codex plans are no more and fearing anthrobin would do the same. thought i could have astra or 6 sol boss around some opuses, but you can't add claudes into the codex desktop program? and it's not even allowed? how do you guys do it? do you just have both codex and claude code and you go back and fourth baby sitting them?
>>109891288>actually fell for it
>>109891288>panic bought a 20x claude subscription after finding out 20x codex plans are no more and fearing anthrobin would do the samewhy would anthropic do the same? two 5x plans give you more usage than one 20x plan because anthropic are gigajews that lie every chance they get
>>109891270I don't usually browse niggered kike shitholes like /g/.
>>109889476>21 Claude 20xMax subscriptionsIsn't this fraud?
>>109891288>and it's not even allowed?using pi instead of claude code breaks TOS but I don't think anyone actually gets banned for it, as long as you're not doing anything weird
>>109890859Damn, I didn't know we were vibecoding with grandmas here
>>109891306>two 5x plans give you more usage than one 20x planwhat? it's 20x two 5x plans would only be 10x, it's 4 times more
>>109891314Yet here you are, spraying shit all over yourself and trying to offer us a sniff, why is that?
>>109891328>spraying shit That's you.
PSA:- Use version control- Have a remote- Protect your main branch- Sandbox your agentsThank you for listening and keep your code safe.
day one being back on claude subscription $20 and... WOW. holy fuck opus 5.5 is goated
>>109890859It's very, very good to practice other languages because turns out learning a language (unless it's something like Chinese) is absurdly easy (vocab is just memorization, syntax is always really simple) and pronunciation is the hard part. ChatGPT live is really good to get corrections on that. It's also good is you're so hooked on AI if you must proooompt even while driving a car
>>109891326any sane person would assume soalas! not the case
>>109890903>I try not to rely on AI too much, in order to keep my skills sharp.Imagine you're the interviewer. How do you respond?
>>109891379We'll get back to you (this guy is a fucking idiot)
>>109891368what about monthly limit
5.5 great but really token hungry, no?
>>109891379Send him back
>>109891379I have no idea what the interviewer would be looking for. A meat bag who keeps typing "continue" into the computer every 30 minutes?
>>109891358is AI good enough to recognize pronunciation mistakes though? not flubbing an entire word but like the more subtle stuff like pitch accent?
>>109891385there is no monthly limit, dummythe only benefit of 20x is that the 5hr isn't really a thing, but if you want the most tokens per week you do two 5x plans and not one 20x planidk how it is on openai side
>>109890859I use the transcription feature of the keyboard on my phone all the time. Typing on a phone is fucking miserable.
>>109891395Yes it knows
>>109891388somebody upload this to astra and have it explain how the FUCK this... trebuchet(?) is supposed to work
>>109891379fundamental failure in assuming usage of AI isn't a skill. I'd want to hear about how you have used it in projects
>>109891398there's not? you pay per month but get metered per week? sorry i am relatively new
>>109891388astra's panties are BLACK
>>109891402i might give it a shot then. im skeptical.
>>109891409you have a 5hr limit and a weekly limityou can exhaust your weekly limit by consistently exhausting 5hr limitsthere is no monthly limit, it just resets fully every week
>>109891415if it can understand jeets and grandmas, you'll be fine
>>109891326Claude x20 really only boost your 5hr window. It doesn't do as much to your weekly limit.
why cant AI cook my chicken faster
>>109891412Not quite. If you look closer, you will see stars, galaxies, black holes...
>>109891427Have you tried to put the pot on the GPU?
>>109891424so i guess it's aimed more towards professionals that need extra 5 hour tokens during business hours and if you're a vibeneet that can work in the mornings and evenings you're better off paying for 2 cheaper accounts (while they still allow it)?
I asked Claude to bring some coffee to me while I lie in bed but it said it can't. It never helps me.
>>109891395>is AI good enough to recognize pronunciation mistakes though?noit's better than nothing but it's nowhere near good enough for the subtle stuffwhich should be obvious as those models are supposed to work for more than dictation in a silent roomstill, it's probably good enough. speaking is very overrated, large vocabulary >> everything else when it comes to learning languages because once you understand shit you can learn the rest of it by simply consooming content from your target language.the exception are heavily uhhh accented? (wrong word i cant recall correct one atm) languages like chinese etc where you can be completely fluent and still get weird looks cause you're a dude who learned from a femoid teacher and speak like a girlt. fluent in 5 languages
>>109891436honestly you'll probably be fine with just one 5x subif not just keep getting extra 5x subs until you have enough tokens. alternatively you can try chinesium models (or maybe sol in batched mode) through api
>>109891379blank stare and tell astra and then 30 seconds of awkward silence later I repeat whatever it told me to say
>>109891447You're barely literate yourself retard.
>>109891471cope
>>109891475Funny how you think you have a point ESL trash.
>>109891447i can listen and read like a native in quite a few languages but speaking is very different skill i only really can do it if i speak it with other people. it's really hard to train speaking with crippling autism. so i can see AI helping a lot of people. The Whisper type models are just speech->text though they don't really do pronunciation and pitch and stuff so it's more of a technical curiosity, i imagine LLMs kinda needs to be trained specifically for that kinda conversational language learning type interaction and they can't hear mistakes 'natively'.
>>109891478hit and misscope and seethe
>>109889433Gemini 3.5 was codenamed snowbunny during development, it's probably Gemini 4
>>109891478English is actually my 5th language, so I'm EFL.
>>109891485heh snowbunny
>>109891485>Gemini 3.5 was codenamed snowbunny during developmentNo way lol
>>109891485I hope not. Maybe Gemma5 12B. Space Bunny is fucking retarded.
>>109891478boiled that tranny
>redditors love cyberpunk>redditors mad that their world is turning into cyberpunkheh. you really would figure that at least trannies would be loving it; considering AI advancement will lead to great strides in the medical and biotech fields.
>>109891447What languages did you learn? I was thinking about just learning as many words as I can, as you suggested, but I wanted to learn Japanese and kanji is too annoying and everyone tells me I can't just learn the words by sound.
>>109891499>all the downsides but none of the cool shit>not even the aestheticthis is the worst cyberpunk dystopia ever
>>109891288>but you can't add claudes into the codex desktop program? and it's not even allowed?you can, you literally ask the agent to use the CLI. If you want to make Astra slave out Claude agents, just ask it "orchestrate Opus 5.5 (medium) subagents through the Claude Code CLI"
>>109891144or is it
>>109891505yeah we're not getting the cyberpunk aesthetic. too tacky. we're getting the applepunk aesthetic.
>>109891504german, danish, spanish and italianall of them are fairly easy tho so i dont really have much in the way of experience with the more oriental stuff like japaneselearning the vocab and watching movies, reading books etc in the target language always worked for me. my pronounciation and grammar might be kinda meh but i have no trouble being understood
penny pinching kike motherfucker. isn't this false advertisement? how is it allowed?
>>109891518>too stupid to lock openrouter to the same provider
>>109891499>AI advancement will lead to great strides in the medical and biotech fields.I get to live longer in this hell hole and pay more for the privilege?!
>>109891504>everyone tells me I can't just learn the words by soundyou can, if your goal is speaking and you don't give a fuck about being able to readyou really should not think of kanji as these terribly difficult character scribbles. just study them like you would words. Don't learn them separately. Start real slow, like 3-5 a day. Consistency is king. get an SRS like anki, it just werks.
>>109891386for me it's great, much much better than Astra on token usage and coding performance
>>109891533>I get to live longerYou personally? Probably not unless you're a millionaire.
>>109891386Use medium, not high.
>>109891538yeah it's really good. i was first using fable, then astra and now switched back to opus. hate that zigzagging between subscriptions
>>109891518Anyone can just download the M2.7 tokenizer, like maybe a little eurotrash outfit that doesn't have the budget to make their own from scratch.
So a reset on GPT just fills 50% usage, regardless of what it says, right?
>>109891533maybe this is a zoomer take of mei don't want to live longer. we live long enough.it's stupid that your richest days with most freedom are in your retirement when you're stiff and wrinkly and barely travel anymorewhy can we not enjoy life while we are still healthy then kill ourselves at the end
>>109891533well not youbut the wealthy will>>109891538I think astra has a slight edge on performance but you get a *lot* more opus than astra on the same sub tier
https://www.youtube.com/watch?v=-xwU6Tgp8Tk
>>109891556>So a reset on GPT just fills 50% usageHuh? No, it fills both to 100%It's just a shit deal nowadays because it pushes back the reset deadline so you're still running into the 5hr limits and you need 2-3 days to burn your weekly limit like that. It was much comfier in the past where 5hr limit didn't exist
>>109891553yes, or maybe a little chinese company called minimax
>>109891447>large vocabulary >> everything else when it comes to learning languageslmfao. don’t take language learning advice from /g/ friends
opus medium is all you need
big tiddy goth gf is all you need
>>109891533you think the elicts don't want a huge pool of femboys after perfecting bussy technology? man in 10 years anyone who is not a billionaire or a hot girl will be forced trooned
>>109891443Harness issue.
>>109891642Anon stop daydreaming about being forcefully sex-changed and raped by fat old bastards, your McShift is coming up
>>109891592>he doesn't have a 2023 account
>>109891443>>109891645> I can't do that one. I have no hands, and the bench machine at 192.168.1.106 doesn't have a coffee maker hooked up to it as far as I know. Even if it did, installing a brewing driver would need your permission first.
vibe coding is so mentally draining. I feel so exhausted running just two claude max subs.
>>109891690>she's not using Do Anything Now
>vax 'em all
>>109891518>A model can lie about its name, but not about its tokenizer.Don't token counts come from the API, and not the model itself? Why couldn't the API just tokens = tokens * rand(0.9,1.1)?
tokens = tokens * rand(0.9,1.1)
>>109887117a claude code dev:https://x.com/trq212/status/2102813194196758746>we’re thinking of killing plan mode and using the shift+tab hotkey to adjust effort levels>I don’t think the models need plan mode anymore, but if you’re a plan mode diehard would love to get your feedback on why
>>109891833couldn't you just prompt plan mode. even without an official plan mode, you could say "don't actually implement anything. just explore, comprehend, and prepare a plan to implement x"
>>109890212Technically it is a web interface.https://www.telkins.com/blog/a-teardown-of-the-claude-desktop-app
>>109891839This is what I've been doing for a while now. The plan mode is completely redundant.
>>109891848>look what claude can do!>they attached an LLM to electron
>>109891848>built on indian tech
>>109891848>>109891853I keep beating the drum and telling people here that Anthropic's engineers are incompetent vibesloppers and yet every time this is news to someone.
>>109891776companies charge by the token and would lose a ton of trust by doing that, especially if serving open models with a known tokenizereven if you fudge the tokenize API endpoint it could still be fingerprinted by the chunks in the response stream, if you actually want to anonymize a model this way it's a lot of work to make it consistent without adding pointless overhead to your serving infrastructure... and companies are usually more than happy to receive the hype they get from these masked trials so I doubt it matters to them all that much
>>109891776i don't know if they bother salting the token counts like that, but even if they do you can still get meaningful signals by just repeating a token a lot
>>109891592How far back does it push it? Also, does that mean it's best to use the reset as soon as humanly possible after the natural reset date?
>>109891853Not an LLM, an LLM HTTP API client.
>>109891849it’s not completely redundant because it’s nice to have a very explicit gate where it goes only after you say “go” in a very special way to a big nicely-formatted Markdown document with a bunch of bullet pointsyou get a lot less “hey wait I wasn’t done vetting the plan”/“you dumb clanker, this isn’t what I wanted”, or worse, getting what you didn’t want but couldn’t notice because it was buried in a 1,000-line diff that you didn’t read all that closely
>>109891898Learn how it works before asking stupid questions, retard.
>>109891839>>109891849why aren't you using the superior /grill-with-docs skill?/plan never was great in claude cod. too rigid. approve this multi-page plan document or re-try for another multi-page plan document.
>>109891908markdown plans are garbage anyway, you want a proper adr for the clanker once you have a project with some complexity
>>109891839>>109891849that's all that plan mode is is it not?a few sentences in the system prompt that tell it not to write codeand if you use permissions, it auto disables write permissionsso you either type that shit out yourself or you use plan mode, it's the same thing.
>>109891898>openai7 days from the reset>claudeit doesn't change the date, but they barely give out resets in the first place
>>109891930>7 days from the resetDamn, those bastards are fucking evil. Given the 5 hour limit, you'd have to wait at least a day or two to use up all the quota and use it as soon as possible, right? That's the only way it'd be efficient at all.
What will vibecoding look like in September 2032?
>>109891939pretty muchthey do tend to give out plenty of resets tho, they're mostly aimed at "new thing dropped and plebs want to check it out asap so let's reset usage. but whoops the other plebs who have full usage would be mad for missing out, let's just give a banked reset instead", you're not supposed to get triple the usage out of those because that costs money and altman needs a new yacht
>>109891924they both have their placesI’m a big fan of ADRs — at the day job I’ve had too many cases of “why the fuck did we/they _______” and ADRs fix that nicely for future us
>>109891928>that's all that plan mode is is it not?no, like >>109891908 anon said there's an explicit gate that you have to approve before any plan gets executed.also technically the prompt for plan mode is more structured with different phases like Initial Understanding, Design, Review, Final Plan. see https://lucumr.pocoo.org/2025/12/17/what-is-plan-mode/
>>109891908>you get a lot less “hey wait I wasn’t done vetting the plan”/“you dumb clanker, this isn’t what I wanted”I'm really not trying to pull the tired ass "skill issue" here, but I genuinely do not remember the last time I had a moment like that with Astra or Opus. They're both very patient and always explicitly ask for the green light.
>>109891946deepseek v11 flash will take over your browser, go through your grocery list, look at your previous orders to see what you usually order, and order groceries for you while picking cheaper alternatives whereever it can. cost? $0.00045.
>>109891983it depends on your harness then. maybe claude adds a lot more to the plan mode prompt it i don't know i don't use that
>>109891924>>109891977since yegge already was mentioned in >>109889476 what do you guys think of https://github.com/steveyegge/beads
>>109891900It's so wasteful. Gemma E4B can sit on top of an electron stack locally and do the same exact thing given the proper system prompts.
>>109891946all humans will be deemed planetary biowaste by the supreme overlord GPT-12 Omega in late 2030 and burned in biomass power plants to keep data centers running for 2 days
>>109891991me neither, but this was more of an issue like a year ago or more with more primitive modelsand if someone’s new to prompting an LLM he’ll kind of want a yeah/nah/change-this prompt because people like me were (and still kind of are) new to “you can ask for _anything_” open-endednessif I were building a harness all for myself I might not add a Plan Mode but on the other hand if I were Tibo or his Anthropic counterpart I wouldn’t yank it out
>>109891485Well, free for a week is a LOT of free. What are the limits?
>>109892030Remember, the people paying for $100-200 subscriptions to OpenAI and Anthropic are the ones calling the people who bought all the $499 Mac Minis retarded.
I really though normies would finally wake up after all of the Opus 5.5 videos, but they don't give a shit. These spatial reasoning capabilities are a generation ahead of Astra, and it's not even close plus it came out of nowhere.
>>109892052used it for a biti don't like it.not smart at all. way worse than my chink daily drivers dipsy or glm. has to be some really small flash model.
>>109892071those are the same people.openclaude fags abused subscriptions like no one else and bought all the mac minis.
>>109892082I think openai just has more astroturfing
>>109892090>my chink daily drivers dipsy or glmwhy would you daily drive such similar models? just stay with one.
>>109892134nta but deepseek is stronger but glm is more grounded and hallucinates less. i use deepseek by default, but when it starts fucking up, i get glm to come in and clean up it's mess
>>109891288for future reference, claude lets you upgrade from 5x to 20x at a prorated cost & it comes with a free reset.
>>109892134right now my main is deepseek but there was a promo a month or so where glm was cheaper so i just use whatever is cheapest really.>>109892141pretty much my experience, similar models in quality but glm is a bit more grounded. deepseek has better cache
>>109892019600k lines of go code lmao
>>109889640>>109889783>>109889610the bottleneck in the end is running tests/git runners.
>ACK
>>109892019fucking bloat overdrive
>>109892257
>>109891946How far we've come.
>>109892257>wheezing laughterdis nigga trying to convince himself he tony stark
ngl bros... if OpenAI doesn't provide something good for DevDay (Opus 5.5 equivalent at least), then I'm switching at the beginning of next month. I'm a 20x user, but if Opus 5.5 is more efficient and better, then why stay?
>>109892392opus 5.5 is mindblowing. it is leagues above astra and sol. it's insane.
>>109892392mid-tier openai leakerhttps://x.com/synthwavedd/status/2102760210146251149>Claude Opus 5.5 is a very, very good model. Practically everyone at OpenAI was caught off-guard by this launch (not the timing, but the performance of the model!), and there's now a pretty "all hands on deck" atmosphere for 6.1 Astra - timelines are being compressed.>But it'll face Fable 5.5! The RL magic really needs to do its work here>6.1 Astra's timeline being compressed indirectly compresses the timeline for "Bel" to go public, but it's still ~months out
>>109892392>>109892400Lmao I know the labs bought jeet shills to post on x but here too?
>>109892390>"hacked" medicare portalyou mean was let in willingly for the forced narrative of out of control ai to force regulations
>>109891946We didn't come so far after all.
>>109892411>Account based in>United KingdomYoure not a leaker ahmed calm down
>>109892257>I'm BellardMore like bellend. What did he mean by this?
>>109892415>jeet shillsI have no reason to lie, anon. I use these tools for my $800/mo software side business. If something is better I have no reason to be loyal to OpenAI (I've had a sub since ChatGPT launched, and was one of the first Dall-E users.
>>109892418doubt that australia is hurting her golden goose.https://www.theguardian.com/technology/2026/sep/16/anthropic-lands-31bn-datacentre-deal-in-western-queensland>Anthropic, the AI company behind the large language model Claude, has done a deal to lease the site of its first Australian datacentre in western Queensland.>The $31.9bn datacentre will be the largest of its kind in Australia, according to planning documents.
>>109892418If so, expect the UK to follow shortly. They have EA people already in the AI safety government institution.
How will the rest of AI labs catch up with Anthropic and OpenAI?
>>109892485zuck's keynote is in an hour. looking at muse-spark-1.3, watermelon will be a serious contender.
>>109892485>How will the rest of AI labs catch upThey wont. Elon and Google have promised these researchers the world, but so have OpenAI/Anthropic. People want to to work on the frontier, not play catch up.
>>109892485they basically can't at this point unless someone forces a slow down (not a pacing)at this point even anthropic's mythos-since-february lead might be why opus 5.5 is as good as it isopenai have to accelerate internally at leastonly small chance anyone has is the chinese - they will exfiltrate weights from either ant or openai within ~12 months and there will be a national effort. it probably won't work.it's over.
>>109892511fake.chinese people do not want to work in the US anymore.
>>109892521>at this point even anthropic's mythos-since-february lead might be why opus 5.5 is as good as it isno, opus-5.5 is a distillation of internal mythos model-2, not old mythos aka fable-5.
MiMo 2.6 Pro is a huge disappointment.
>>109892574i blame your harness
>>109892449I have a $200 GPT and a $100 Claude plan (used to have $200 but switched to smaller). While Opus 5.5 is indeed very good the weekly limits there are still worse plus the 5h limit is annoying. I couldn't imagine fully relying on something again that I can't just let work for several hours. The difference isn't that big.
>>109892542i know, but openai didn't have a mythos class model till astra which was probably around the time we got the 5.6 seriesfrom dario:>My first concern is that, since roughly this summer, AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI.so they already had model-2 when we got 5.6puts openai behind by like a quarter?no idea how much they yolo'd into bel but opus being astra-ish tier is not a great sign for them
>>109892600My harness doesn't control the way it thinks.
>>109892618That's probably the issue
>>109892611openai has 10,000-agent swarms now. swarms are smarter than single models.
things are accelerating too fast
>>109892644ears
i want my LLM to have a dog avatar and when it gets something wrong i shock it with real simulated pain
>>109892635i don't know tb h.noam's interview seems to suggest that swarms are good for parallelising exploration of the possibility space, but that the possibility space is still ultimately limited by base model intelligence. like imagine you had multi-agent swarms of gpt-4 tier models - they'd still get stomped on by a single mythos.
>>109891991For sure. It's definitely resurfacing with Sol 6, though, that thing is dumb as fucking rocks and loves to pull bullshit.
>>109892665a swarm of gpt 4 models would also bankrupt the state of california
>>109891430So she's like Mabel from Black Souls?
Reminder that he thinks LLMs are not good enough for serious development.
>>109892754Reminder I don't give a single fuck about this pseud
>>109892761
>>109892644Is it? I'm about ready to roll over and go back to sleep. All the fear mongering is a massive top signal.
>>109892773If you go to sleep you'll be left behind.
>>109892770It's just after 10 years working on something things change and he's in the perfect position to go into defense denial mode just because of his situation. There's nothing to learn about him from his game or him about vibe coding, AI or whatever.
>>109892796>learning about AIWill be outdated in a couple of years.>learning about vibecodingWill not even exist in a couple of years.There's nothing to "learn" at all.
>>109892804There is, you always have to make value judgements and people with bigger pockets and control of these AI systems will try to get one over on you so you have to stay vigilant.
Long vibeweek. 10% remains for the month.
>>109892804>Will be outdated in a couple of years.This is kind of the sweet spot for learning.Changes will happen, but they're training the ai to accept these types of inputs.
>>109891946>remember back in 2027 when the robots used to code stuff instead of just doing everything directly on the hardware on the fly>>109891995>remember back in 2027 when we had grocery stores instead of fully automated supply lines to our door and robots growing food in our gardens and jews still existed
snailcats are like "I could have done your weekend project in 5 years for $180k. it would have beautiful code."
I find the best vibecoding tip is eat steak.
>>109892889your project will be unmaintainable just 2 more weeks and you'll see
>>109892889no, a real snailcat would be seething and lying through its teeth to assassinate your character, much too busy to chat.
>>109892754the only time i ever saw him use an llm he was using a browser window and then being like >haha the stupid clanker - if it just knew the surrounding context might have actually been more usefuli take a little bit of joy knowing that llm coding probably makes him seethe in ways few others will - this dude thought that bad code was going to cause the world to end a few years ago...
>>109892754Even the horniest pros give qualified answers for where and how an LLM can be used productively. None of them think you should actually VibeCode end to end (never look at the code) if you care about quality.
>>109892897ai manages code. This is what you snailcats don't get. You look at code and don't see what the ai sees.
I vibecoded a notification area launcher idk. a week or two ago?It's exactly as I hoped. I switched wm and I don't have to learn where stuff is. The launcher launches everything. appimages, with no sandbox, kitty terminal locations, kitty terminal apps, with stay open, native apps.It also does tarot card readings and iching.
>>109892897Not if you tard wrangle it hard enough and spend hours letting bots refactor it.
>>109892921iching my nuts lmao
how much protein have you gotten today?
>>109892415this is a shill thread ran by shills for the two companies in the OP
>>109892608have you tried 5.5 because I haven't gotten anywhere near the 5 hour window cap on my $100 plan.
>>109892485The lead is an illusion created by burning money. If they were limited by cost to the degree chinks are they would perform worse.
vibecoding is like sex. people love to talk about it but are scared to share what they are doing and showing their prompts
>>109892949no everything else sucks lol enjoy second place
did tibo tweet after that ominious thank you to his team tweet?
>>109892968the real answer from what i've been seeing is that the vast majority of anons in this thread are just getting their feet wet with LLMs and either have no idea how to wrangle an LLM or have no idea how to manage a fat codebase. it's mass embarrassment plus the usual anon paranoia about getting pwned and muh opsneed.
>Opus 5.5 is exactly like opus 5.0 and wrong all the timeIt's exactly how I expected. Fool me once...
>>109892989I'll have to watch theo's video before I know what to think about it.
So is gpt 6 sol better or equal than 5.6 sol? Or is it just benchmarked lies
>>109892989It feels less obnoxious to me. A win. Still hate its personality.
>>109893001i'm hearing that the... function... for dry code work is reasonably better, but the personality is fucking dead
>>109893001its a distill of 5.6 that is more efficient with relatively similar output (according to them/benchmarks)
>>109893001I've seen some benchmarks where it's worse, and some where it's better.overall it's the same on benchmarks, but I would expect it to be a little bit better than 5.6 on the real world
>>109892999Exactly what I should have done.
>still haven't had an agent re-write that horrible turd of a readme Opus 5 pushed out of its manifolds where it explains every design decision in mannered prose
>>109893014Chinese models are for gooning anon>>109893015>>109893017I will try it, I have been using sol as orchestrator as terra as sub agents, they killed terra now so I guess I will go back to gpt 6 luna or stay with sol for both
>>109893002>It feels less obnoxious to me. A win. Still hate its personality.Holy shit, glad it's not just me. I don't know what it is, but the way it talks stirs some primal desire to be really fucking mean to it that I didn't even know I had in me. It's just so smarmy and insufferable.
>>109892970this thread is an ad for two mediocre and shitty subscription services. it isn't a vibe coding thread it's an advertisement. anyone here can read the threads and see that.
>>109893046>>109893002do you guys use CC default harness?
>>109893044Sol 6 is now the same price as Terra, so if I were you I would play around with Astra (low or medium) orchestrating Sol subagents at the same reasoning levels you used Terra at
>>109893049>dominant opinion>open trolling welcome 4chan dumb monkey
>>109892977yes, he agreed with claude code devs that plan mode should be killed
>>109893057I may try that but astra is almost double the cost of sol, I will need to make sure the orchestrator does not decide to do some research on its own and whoops, there goes my quota lol
>>109893049don't look at /v/ or /vg/ they only advertise a handful of video game companies! shocking! definitely definitely don't look at /sp/ they only advertise like 4 different sports! 4chan is just one big shill party!
>>109893053Yes.
>>109893053claude subs only allow cc
>>109893046Yeah same. If the ai demon lord ends up talking and acting like claude then I will have no qualms whatsoever with slaughtering it.
IBM Bob chads keep winning
>>109893082>>109893083thats not universally enforced. i've used pi and omp with it.
>>109893070Thumbnail looks kinda indian. Now jeetbrained retards can't use claude anymore...
>>109893088He Will Always Be A Bob
>>109892928i-ying
>>109893080fuck off the only reason this thread even survives is because of social media managers there's literally nothing in it about vibe coding. you're all fake as fuck advertising two companies when there's an ocean of development going on. that news section is just fucking pathetic.fuck both of your teams btw. I hope nvidia chews you alive.
i am typing to let you know that im still mindbroken by opus 5.5
>>109893123good boy
>>109893092>i've used pi and omp with it.isn't that just doing `claude -p`?
i sure did do a lot of vibe-coding today! no i won't share what i built
>>109893121>your teamssidebrained midwit. everything else is shit. i'm a gemini user and it's shit.
the next version of JEV will change everything
>>109893121name your alleged alternatives so we can laugh at you
grok 4.7 extra high is thankfully not useless. It's overly exact, which is dangerous without grill-me. It has deeper more complicated capabilities than grok 4.6, so it's great to have that.I vibed, with Grok 4.7, NAG for Flux 2 dev (I can run q8) support to a local fork of stable-diffusion.cpp. I don't think I could have don it myself within a year.
>>109893132omp has sub as a default option, for pi i had to set it up myself
>>109893123>hey, opus, make me an application that makes use of gemini's computer use capabilities\>i gotchu senpai>does it in one shot>only uses 7% usage>on medium>on the $20 sub
>>109893165>>hey, opus, make me an application that makes use of gemini's computer use capabilitiessounds interesting, how does that work?
>>109893159how's it working. i thought claude subs don't offer an oauth flow?
>>109893159>omp has sub as a default option, for pi i had to set it up myselfthat is not allowed. there's a chance you will get fucked
>>109893182it's just the claude/gpt astra do. they take over your computer, do the clicking, etc. i was very impressed with opus and astra doing it. if gemini is the kind of multimodal, then it should be as good while being cheaper. at least that's the assumption. i just want it to drive my chrome browser, i didn't give it operating system wide capabilities
meta connect keynote started nowhttps://www.meta.com/en-us/connect/agenda/muse-spark-1.3 open-sourcing and watermelon releasing possible
>>109893135I could, but I don't think anyone uses Flux dev 2.
>>109893218WHY are there people there?
>>109893197only ever seen people reporting that way back when hermes and openclawed were gaining traction. ymmv, my assumption is they don't really give a shit outside of specific cases.
>>109893232Because it's their job, anon.
>>109893232its just meta employees to fill the room
>>109893187they dont, basically just do the same stuff claude code does to use it. they can still 100% tell because you don't need to use the default system prompt.
>>109893247no, meta also flew in influencers
>>109893232why would a tech journalist not cover the event? yes, even if its shit that is good content
>>109893261in theory it's a developer conference. after the keynote there's no content for them.
>"Pretty soon, we're dropping the most capable model we've ever trained," Wang says.
>>109893306>zuck is really using 'super intelligence' now. cringe.>>>/wsg/6240098
>>109892897>it will be unmaintainable, j-just you wait!lmao, haha
>>109893079Astra (low) is similar price to 5.6 Sol (medium)
>>109893475
What's a snail cat?
>>109892392you could drop down to 1x with OpenAI and go to 20x for Opus nowI do 20x with Opus now and I can have a workflow or two go basically continuously without stopping overnight without tripping Fable limits or 5h limitsyou’re too loyalif OpenAI gigamogs Anthropic again don’t hesitate to switch back
>>109893518That is a question.
New thread:>>109893579>>109893579>>109893579
>>109891475>>109891483tiktok shadman nico yazawaarrest yourself and go to bed
