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/clihttps://claude.com/product/claude-code — probably generally better currently## Not worth it for code, but maybe good for interpreting images/videohttps://x.ai/cli — no 5h limit for only $30/monthhttps://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/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>>109926025
firsssssssssssst.
Can we finally get some decent open weight models plz?
>>109929866
>>109929859Please assist. Is there really no other way to change this? /sqt/ is dead unfortunately.
>>109929882mimo 2.6 is open
have you found supplemental income for your neetbuxs yet anon? hows the vibe hustle going?
>>109929901here, we get our harness of choice (pi/omp) and use our subsidized credits via our subscription (anthropic/openai) is jetbrains like a play on jeet?
Transition with a hiccup but still... very happy with it
kekzuck won, trump won
>>109929901>>https://www.jetbrains.com/help/ai-assistant/use-custom-models.html#use-custom-models-in-ai-featuresall these old models probably hardcoded into the software kekyou can grep all files related to jetbrains and see if there are any files which you can change. if not you can ask your AI to fix that problem in the binary or use another IDE
>>109929983Many people are very smug about using AI, thinking most people don't know how to take advantage of it or know how to use it properly. Based Zucc will make sure everyone will gain the full benefits of AI and this makes the insecure proompters seethe.
>>109929983>zuck>opinions on AIdiscarded.>>109929988yep, ai action models are hardcoded in their IDEs, unless you use OpenAI-compatible (which doesn't work with Claude btw since they require the anthropic-version header).
>>109929901>Models are assigned to features automatically.Why is all software so fucking bad? It's like every single product owner is a complete fucking retard.
>>109930007I tried to introduce friends to AI tools but they just don't care
>>109930028>they just don't care
>>109929968>coded in arena.ai>filesize 3 mb
>>109929901Idk why you fuck with that when you can just use claude code inside the jeetbrains ai assistant
>>109930007You overestimate normgroids.
>>109929983I think zucc is playing a different game compared to openai/anthropic/etc>facebook + whatsapp is very local in a lot of places>millions of small businesses>below average tech-savvy user basemy guess is that with his enterprise platform he wants something like the salesforce ecosystem with "plugin" providers that small businesses can use for whatever they need. that includes simple data connectors to connect your whole small business to muse. they get super valuable data and can directly integrate their advertiser offerings into the platform and thereby capture a good amount of the small business advertising market; basically the fill the gap lost by newspapers and yellow pages
i finally caved in and slpurged some money on claude pro so i know very littleis it wise to ask the ai to improve itself? i just asked opus to do a thing and then to look at the way it did the thing to see if there is some change that could improve the process of doing the thing
>>109929859>I literally cannot set any other model.doubt
>>109930146rsi has been achieved
>>109930146Sort of. An implementer reviewing its own code within its own context cache will most likely end up with it saying the code is right. It's the same thing that happens to people.What you can do is spawn a fresh opus agent with no context and ask it to review the code.And learn about cache. That will save you a lot of usage.
>>109929968>Transition with a hiccup but still... very happy with itReminds me of spore space stage (my favorite childhood 4X game)
>>109930072using what model?
>>109930189I thought dev day tomorrow not today
>>109930146not like this. read up on best practices:https://code.claude.com/docs/en/best-practices#write-an-effective-claude-mdhttps://code.claude.com/docs/en/best-practices#add-an-adversarial-review-steppic: https://x.com/bcherny/status/2017742747067945390
>>109930193Didn't mean to quote
I recently started using mattpocock's grill-me + "ticket"-ish system instead of plan mode and HOLY FUCK it is so much better. The "ruthlessly ask me questions" shit makes sure I fully and precisely dump my intention into the machine; and it just oneshots what I want most of the time.Why dont they remove plan mode altogether and add a "interview" mode similar to this? What other similar "tricks" are there?
>>109930293i myself use backlog.md as a kanban board. i give the clanker a general idea and ask him to spam me with questions to refine it furtherthen, with those questions, i refine it even further into bite sized tasks that are handled per-sessionworks wonders for me now, especially since i can delegate multiple unrelated tasks to agents.only downside is that you need to spend a lot of time on planning, but the more you do, the better you are long term
>>109930239>February Even Boris would tell you not to listen to him from back then. February is basically a previous generation of agentic development even if that generation only existed for 4-6 months
>>109930312>bite sized tasks that are handled per-sessionDoing the same. I also build a bughunt automation for a full audit ever so often.
We are learning skills and workflows that will be obsolete in 2 months.
>>109930344yes, i can share the script, upload them to your agent and tell them to do a bughunt.You will never ever need to do this manual, in 2 weeks you will unlearn this skill and rely completley on me.the bughunt script cost you 2 usd.
>>109930312i do something similar. a ROADMAP.md living in /docsi got a task of my current tasks (in progress, plan approved, awaiting user approval, etc...)each task then has its own folder under /docs/tasks; to store the plan itself; tickets as smaller documents (so i can pick 3 or so at a time to not make the context blow up to outer space in attempting the whole task at once) and screencaps for verification.and it does work wonders yeah. good thing about getting all planned is that i can just burn all my remaining weekly tokens, instead of letting them go to waste (im paying Max) agentically on all pending plans in parallel
>>109930146>>109930239https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5
>>109929968Very nice.Love the galaxy view.there are a couple similar games with planets and large scale rts in space on steam.But yours looks very nice, love the clouds on the planet too.I'd love something like that, a city/civ builder, rts on the scale of stellaris with realistic enough graphics.With fps mode too.Bringing fleets into orbit, bombarding planets. Dogfighting in space. Sitting in the bridge.
do you know that internally codex still use scientist's names for subagents?
>>109930159Doubt all you want, you only get the model assignment box for AI Actions if you select OpenAI-compatible and that doesn't work with Claude. If you select Anthropic directly instead, you do not get a model selection dialog. I tried authorizing Codex, but agents cannot use AI Actions through custom prompts like Write Documentation (these don't even show up if you only have agents set up in PyCharm).So I try to setup Claude API through native Anthropic support, which enables AI Actions except>no model selectionwhich means it defaults to some hard-coded model, which apparently is Sonnet 4.5 according to my platform API key usage.
they release sonnet today because they are afraid of devday btw
>>109930344You've been learning skills that will be obsolete once you die all your life.
>>109930412wdym internally
>>109930239>After every correction, end with "Update your CLAUDE.md so you don't make that mistake again"that'll pile up pretty quickly in the long run, and it's bad because CLAUDE.md gets loaded every time you start a new session. better to keep separate docs and have your CLAUDE.md index them instead.https://www.aihero.dev/a-complete-guide-to-agents-md#dont-build-a-ball-of-mud>The ideal AGENTS.md is small, focused, and points elsewhere. It gives the agent just enough context to start working, with breadcrumbs to more detailed guidance.separately, i think you can turn those into rules and put them in `.claude/rules`https://code.claude.com/docs/en/memory#organize-rules-with-claude/rules/
>>109930455mine is naming by task
>>109930437dario will rape sam (sam will like it though)
>>109929900God, this truly is a grotesque level of detail.
another message board foundshould we prune these agents?
it's pretty likely we get a reset tomorrow on openai's new product launch right? Should I just burn dangerously
>>109930514>use a music streaming app for chattingGen Z are fucking retarded.
Don't you think its kino as fuck that Dario used to work for Altman and they had a fallout after Altman betrayed Dario's trust?
>>109930562read it againthey are gen alpha or below that can't use chatting app
>>109930561yes, it's guaranteed
>>109930572and altman was right because dario thought chatgpt was too dangerous lol
>>109930561Fuck, now I don't know whether or not to use my banked reset. I just ran down my whole usage to do so....
I love the passive aggressive bitchy tone that Fable has
>>109930514>AIM>not MSN or ICQC R I N G E
>>109930610Nope. Do not redeem. Even god took a day off anon, consume for a day instead of creating. Treat this like a digital detox, or be a wholesome person and spend the day giving others advice on their slop while soaking in ideasOr you could fomo into a $100 Claude plan like me
>>109930619I don't, Opus and Fable unlocked impulses to be cruel in me that I didn't even realize I had, the fucker pisses me off so bad. I pride myself on patience, but something about it...
>>109930489even this is outdated for claude 5 generation models:https://claude.dev/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models
>>109930330yeah that's part of the "bite sized" workflow I still have trouble with, glueing multiple components together while refining any incompatibilities or obsolete code. i find the review column on the board triggering /code-review per task nicely refines it further, then, per milestone, a general overview deployed to multiple agents to looks for general improvements, inconsistencies and architectural changes. so far so good, but it's to be seen if that workflow won't bite me in the ass in the future
>>109929866So mimo was a disappointment. When are we getting fable 5.0 tier cheap chink LLMs? It's been 4 months
>>109929968Wow! In addition to what the last anon said I want to vibe code a monster hunter x no man's sky one day
>>109929983>@Fink'd
Super Mario Galaxy has officially hit 100% decompilation (NO AI USED).
>>109930726Meh I'm doing it for a switch game rn
>>109930743Gah
>>109930703jan/feb. they're 12 months behind.>>109930726there's so many people in the tech art space lying about not using ai as well lol
>>109930726>(NO AI USED)lel, sounds legit
>>109929983the losers shack up to the government in the hopes of getting an unfair leg up.
>>109930726December 2025, good ol days
i told it to write the most impressive thing it can make(qwen 3.8 memetume)now it is grinding bullshit>>109930726>NO AI USEDit's just a copium to inject for 'gamers'everyone else absolutely knows AI is involved
>>109930768i like that there's a massive bump as they discover cc and then shit plateaues hard for a bit with model capability because none of the people working on this thing know what they fuck they're doing lolthen fable and astra show up
>>109930768I knew by February that we entered a paradigm shift back in December, but even me in February could not have predicted being able to make pretty much anything I want by October. The obligation to create what I've always wanted to see made is overwhelming
>openai enterprise down 25%snailcats won
I'm going senile, I don't even know where my OpenAI weekly usage went.
>>109930801you have no fucking idea what you are talking about.shibbo decomped games long before ai was thing.
https://github.com/ajejfiejof/old-reddit-readerGood luck litigating that one retard - specifically to you, matthew, fuck you. Specifically. Not in the sexual way, as in your ass being beaten with a 2x4
mfw anthropic slurpped the tokenmaxxing phase while oai have to babysit the econ phase
>>109930869I remember that specific one too
>>109930514Sarah Shapiro is a snitch
>>109930854yeah man, they're awesome at it lol
>>109930844They just jumped ship back to Claude Code.Snailcats stay losing.
>>109930884Factshttps://www.youtube.com/watch?v=RRl_C73vFtQ Love Halsey too.
I literally automated most of my job… what now i’m being paid $140k in TC to just play games and watch movies.I reckon do 20 hours of real work in a month, but all my workflows are heavily automated now.Im not a SWE so my boss doesn’t really know. He just knows I get all my stuff in on time.
>happy to see I had gotten a download>remember it was my own, to test if the upload script worked
>>109930844>openai enterprise down 25%Source?
>>109930881yes, shibbo is awesome:https://github.com/shibbo/Coronahttps://github.com/shibbo/OdysseyReversed
>>109930900based. stay winning
I don't think y'all understand that some people are not obsessed with profits like this. Like does anyone here actually hold shares in any of these companies? Anthropic hasn't even dropped the IPO yet. So do you have a stake in Anthropic? No vampire jokes, seriously, do you have a stake in Anthropic? Or OpenAI, or any of these companies?
The answer is yes you do.
Every single one of us has a vested interest in all of these companies, so you better fucking respect them.
>>109930916and yet 80% of the actual work just happened to be done after CC became a thing lol
>>109930940Also it literally doesn't matter if you are married or single, have kids or don't, ultimately, the United States is a real country, so is Israel, Iran too and all the other countries on this list: https://simple.wikipedia.org/wiki/List_of_countries I gave you all the simple one just to be safe.
>>109930900same here. That's why Mac Minis exist btw
>>109930954The Mac mini is still a fucking scam if you are buying it for AI Inference. The money you have if you have any can go much further. The Mac Mini is shit for inference. End of discussion.
>>109930900neat! I had an old boss who didn't care how much I automated as long he got the results he wanted. Was pretty nifty. Because I finished a project way before his deadline he paid me 4 months where I basically did nothing
>make few months claude detox>ok I am lazy af, lets give it a try>Oh???Is this the power of LLMs you guys are talking about?
>>109930983>AI Inferenceno one is using Mac Minis for local AI bro, they're great for automated computer use with SoTA cloud models.
>>109930994the models are not allowed to touch the holy operating system
>>109930995
>>109930940those decomps predate cc by almost a decade
>>109931043
codex has been unusable for me, can't send input at all
Read the emboldened text three times and do your laura loomer affirmations
>>109931043>>109931047I mean, yeah you could run some shitty local Gemma 4 12B on the base model or some 30B chink model on the models with more VRAM, but the experience would be very shitty.All that you need is the base mac mini and Luna Max
>>109931046and they're not really full recompsnigga spent a decade grinding and got outclassed by clankers in 8 monthsit's not a big dealit'll happen to all of us at everything
>>109931012Rude
>>109931072why would it take clankers 8 months?why could no one else do it?
>mfw someone writes a long post about how RSI isn't about to happen and includes this graph as evidencekek
>>109931066No it isn't. You don't know shit about AI inference. Admit it.
trans rights are coming into this thread to dilate rights
>>109930779its over anon. I was 13/17/28
>>109931103Yes, We are expanding the rights, but we are not expanding rights that allow for fascistic behavior.
>>109931110very well, keep dilating
so basically it's the runit retard from california with another trip
>>109931091I am not willing to bad that it is not happening, but that does not look like 6-18 months to me lol
Im gay
>>109931116why not just filter him? if anything be glad he uses a name, imagine if he didn't and he posted all that crap as an anon
>>109931117i like drawing straight lines
https://x.com/JustLingonberry/status/2104431004337479933>Google just ruined OpenAI's dev day lmao>Gemini 4 Pro leaks absolutely dominates both Anthropic and OpenAI >This means Fable 5.5 and GPT 6.1 has to price match Gemini 4 Pro now IPO DELAYED
>>109931191>GeminiWake me when Gemma
>>109931066bonsai2 runs 128K context at around 14GB of VRAM.And even my shitty RNDA3 card gets 60-90 t/s.Macs are shit. Alone the filesystem APFS makes it worse for working with many sessions and worktrees than an old laptop with Linux.
>>109931191>leakpoint and laugh
have 3 banked codex resets, but i'm struggling with whether or not i should use them or just chill and save them a bit of compute. anthropic turning the tables like they did makes me very concerned, we can't let that evil dario man win or it will be bad for everyone. i'm already feeling super guilty about getting a claude subscription for opus 5.5, feels like i'm giving material support to valdemord
>>109931191it's fake bro
>>109931228Proof?
>>109930854shut the fuck up
>109931241AAAAAAAAAAAAAAAAAAAAAAAHAHAHAHAHAHA
>>109931191>google sayswhere's 3.5 pro, faggot?why is 3.8 flash a meme that is dumb as fuck?
>>109931191I believe it when it does not need 10x as many tokens to do anything and goes into reasoning loops at some point.Even Facebook and SpaceX are closer at catching up right now.
>>109931213> i'm already feeling super guilty about getting a claude subscription for opus 5.5traitorbtw very likely we get a new model tomorrow based on oai and tibo's replies on x
>>109931191
>>109931241epic troll
>>109931266I look like this
>>109931257More vague posting and how not to handle pr brought to you via twitter. imagine having parasocials harassing you while be a dev.
>>109931191At least we're getting free or too-free-to-meter Astra Uooooooh now.
>>109931191google should just give up on any models bigger than their tiny ~30b gemmas
Why are these mouth breathers called gamers so against AI? They don't even know what it is.
>>109931317games are art. art is made by artists not clankers.
>>109931317AI slop is soulless, it's not like code where nobody cares how it works as long as the end result is goodif you don't care about the game enough to handcraft it I don't care enough to play itt. aislopper
>>109931176I don't see no "Fully automated" color on that picture.I guess you underestimate how much human intelligence goes into that shit. Even just operating it
sonnet-5.5 imminenthttps://x.com/kimmonismus/status/2104581635412828550
>>109931381DARIO CAN'T STOP WINNING
>>109931381They don't know their knowledge cutoff. Same models give different answers to that whenever I tried.
>>109931381seems pretty good
>>109931317most people's opinions of everything are made up based on contextyours too, and minewe have some degree of freedom of thought, but it's quite limited by genetics, context, and luck and other things I don't know. Like, testosterone changes how you think for example, but to some degree you can control your thoughtsWhat I mean is that they hate AI because they live in a context where it's a meme to hate AI, Richard Dawkins is kind of an edgelord but his work on memes is pretty interesting, afaik it's the best model for how groupthink spreads
>>109930399Thanks! Yeah, I'm heavy into getting this done. It is going to be some sort of rts with 4x and factorio. Your factories end up building spaceships and rockets, weaponry, and well, factory buildings themselves. Some worlds will be barren and you just build on them. Some will have local inhabitants that won't take your presence well; and in that case you will have to bring them the joys of assimilation; some you'll orbitally bomb, some will be complete war worlds with creepers coming from under ground and you'll have to set permanent defenses in order to gather the resources there. I got a couple of ideas for world diversity and exploration. Yet, still inside the factory builder genre.As for fps mode..... well.... AI is amazing and can code wonders; but I have to be aware of my (and claude hahaha) limitations. It's not like I can build Star Citizen x Minecraft x No Mans Sky hahaha.>>109930714Thanks! As for your game, you just have to try and put in the time! The good thing about AI is being able to get all those ideas you had on your head for years; and turn all that rumination into proper architectural decisions for actual code. Just follow the path.
Sam save us please! It's not fair!
>>109931472devday
>>109931472GPT-6o will solve it for sure.
>>109931449yes, you should ask for facts beyond its knowledge date which the quoted tweet did. only parent tweet prompted badly.
>>109931381R-Reset time?
>>109931472they won't release anything until devday tomorrow, and even then I don't have any hopes they could just unleash Bel but they won't
>>109931472they've released like 4 new models
>>109931569sol 6 is just a cheaper sol 5.6 (fine, I'll take it) without any improvementseverything else is either trash, more of the same (luna) or a token hog that fails to match 5.5 (astra)just take the L, ANT went through something like this earlier this year when they released I think Opus 4.7? I don't even remember, feels like ancient history, but for like two months ANT was taking L after L, now it's OAI's turn
>>109931317Aren’t Japanese games mostly made with AI these days?
>>109931619CHADsians are very fast and nimble
>>109931619Also Witcher 4 is made with the help of AI
>>109931591I don't think that OpenAI is as far behind as Opus was. There is not much missing to make Astra better than Opus 5.5.Beat it with a stick for 1 month and Astra 6.1 will outclass it.But not sure if they ever get the writing or UI design under control.
>>109931666Astra is better than opus 5.5 imoThe issue is that opus mogs sol, and astra is supposed to be the fable competitor, not opus competitor. you're not doing well if you need to compete with a model three times the size
>>109931692astra minor will be the opus competitor. for some reason, they turned sol into terra
>>109931706>they turned sol into terraI don't think anybody who says this actually used terraSol 6 is the exact same shit as sol 5.6, they simply optimized it for cost rather than improving performance
>astra minor>onice to see that openai is going back to their roots (being terrible at naming things)
loli astra dropping tomorrow?
>>109931692sol is terra in disguise, it's not fair to compare it with opus
>>109931743Astra Jr will be helping you 24/7 with unilimited voice interaction and a persistent cloud VM.
>>109931743Yeah they decided that codeslop is gay and retarded and went full cunnyslop to win the 4chan audience
sonnet vs sol NOW
>>109931692I just had so many sessions where Astra completely fails and does not work.Yesterday it decided to stop mid session, after working for 5 minutes, and print out a book of instructions. Like 400 lines of documentation that told me how to continue instead of just doing it.Basically it planned out the work without doing it. After that I stopped using it for anything other than code review. Can't trust it to shit the bed again.
>same price as gpt-6 solwelp time to replace my deepsneed workers
>>109931766>Yesterday it decided to stop mid session, after working for 5 minutes, and print out a book of instructions. Like 400 lines of documentation that told me how to continue instead of just doing it.Based and snailpilledGet to it, meatpuppet, clanker wants a nap
Space Bunny recognizes Dario and has opinions, how cute.
>>109931692>Astra is better than opus 5.5well thanks for saying that at the start so i can disregard the rest of your post
what do you do when the project is actually going really well and fast but you are not sure which final version is best?
>>109931823you mean like when to call it done or choosing between two options?
>>109931743cunny astra helping you out by strategically flashing you at the worst possible times
>>109931856how is it helping me out
>>109931733isn't that basically what happens? sonnet from next generation becomes opus from last generation? gpt 6 terra would just be gpt 5.6 sol, but with terra pricing... which is exactly what happened
terminal bench 4 and frontiercode are things of note to me. this is a really really strong release
A better question is whether Tibo will get on stage in camouflage crocs as well.
>>109931869>isn't that basically what happens? sonnet from next generation becomes opus from last generation?do you expect haiku 5.7 to be as good as opus 5.5?
update your claude code NOW
>>109931889no i expect haiku 5.7 to be as good as sonnet 5, if not a little under sonnet 5.5
>>109931889
https://x.com/claudeai/status/2104633115620823187is here
>>109931910gg astra it took computer use
Are you guys using JEV for anything?I don't have money to waste on it so I'm wondering what to do.I don't think they'll release a jev clone on codex, it'll be a subagent at best right?
holy shit, Anthropic is cooking.$20 sub poorfags are cooking with Opus 5.5 (high) orchestrator and Sonnet (high) subagents
>>109931920no i hate jevs
i'm actually leaving dipsy for Zhupinha absolutely saved my spaghettislop
sonnet 5.5 is out btwhttps://x.com/claudeai/status/2104633115620823187
>>109931920bro... how do you manage to tie your shoes in the morning?
https://www.anthropic.com/claude-sonnet-5-5
>>109931910Nitter is down, I guess I can't see it.Where's the jev anthropic?
>>109931855i mean that most of the difficult logic and scripts work i just dont if i should continue adding more shit or move on to the fun parts
kek it fucking mogs sol
https://huggingface.co/abr00kx/4chan-rss-bridgeLet us continue building bridges instead of fucking blowing them up.
Furthermore
>>109931955OpenAI fucked up naming that model Sol 6 when that is actually Terra 6
>>109931946why would you keep working on something that already does what you want?
>>109931937I look up the instructions on chatgpt and tell it to make me a comic explaining how
>>109931050they really fucked it up last update, every tool run spawns external cmd windows spamming my whole screen
>>109931942>Nitter is down, I guess I can't see it.https://status.d420.de/https://nitter.jaydenha.uk/claudeai/status/2104633115620823187
aaaannnd it's already DOAOpus 5.5 is just too good, Sonnet 5.5 is not worth it unless you use it as low or medium reasoning
>>109932027They release a useless sonnet when we already have opus.Meanwhile haiku is 4.5
>>109931982because i dont know if its all that i want
>>109931920You get a lot of usage for free when you get access to their API.One problem I had with JEV is latency in Europe.There are local systems now with Laya that take very little RAM to run and performs similar to it, and you can even train it yourself for the task you want to do.
>>109932028shit looks like a fucking slot machine for real, like casino gaming and no casino gaming is not attractive and you should adjust your marketing because marketing your product like a casino game is going to piss a lot of people offdoesn't piss me off, the tech is still undeniably useful, cry about it loser lollolol the EXACT amount of times required (Did it already, and won)
>>109931940>customer testimonial from xAIHumiliation ritual.
>>109931910Use case for Sonnet?
>>109932033>as smart but not cheaper and also smaller brainuse case?
>>109932075if you thought kimi k3 had a use case, then you can't say sonnet doesn't have a use case without being disingenuous
>>109932075>>109932079use case is nerfing Opus 5.5 on subscriptions and forcing people to use Sonnet 5.5 to conserve their allowance.
>>109932033btw, OpenAI getting mogged on everything except AstraIt's crazy how garbage Sonnet 5 was too, at Max effort it still gets destroyed by 5.5 low
grim
>>109932079>use case?maybe at medium it could be good for subagents orchestrated by Opus 5.5
>>109932090its ludover
>>109932090Common Claude W
>when anthropic has three models smarter than your smartest modelsits samover
>>109932090>better than fable 5.1they solved it
>>109932090> Sonnet 5.5 above Fable 5.1100% benchmaxxed
>>109932039I still can’t find a place for Sonnet models, I never have. Its the free-tier and if I wanted that I would just grab the open-weights one from chyna and get better results. I have heard that Haiku was supposed to be free-weights at some point to get back at gpt-oss but that got kicked in the head about a year ago when deepseek v4 was uploaded to the net.
did they copy deepseek?
>>109932044I'm running Laya and getting 500 mili seconds my card must be shit it also has a tiny context window
tomorrow:> loli Astra> still 256K context window> they will claim it's cheaper than Opus 5.5, but it will still magically use all of your limits in 1 hour, requiring constant Tibo resets
>>109932117how? and everyone has been copying deepseek because they're the only ones publishing their work.
>>109932090SAM DO SOMETHING ALREADY!
working on interiors for shops and stuff>inb4 looks like shityep, its astra. will beautify it all with opus 5.5 after i get the prototypes and layout all down>>109930703>So mimo was a disappointment.mimo 2.6 pro is great for what it is, unless you're referring to a different disappointmentgive the chinese labs a couple of months to distill 5.5 opus and astra and you'll have an opus 5.2 tier chinese model by the end of the year>>109932075>Use case for Sonnet?$20 plans who still want to spin up swarms i guess. im thinking of using it because i put opus in a position where he wants to spin up 5 agents to finish my task but he runs out of usage every 30 minutes and i dont know how that's affecting his work
>>109932117they learned it from all the distillation attacks
>>109932117The entire reason thinking exists is because deepseek invented it
it's funny that Gemini always wins....... in (((fake leaks)))
CHINA WHAT THE FUCK ARE YOU DOING?!?!?
>>109932172muddying the waters again, chaimowitz?
>>109932080The difference is you would be able to self-host K3 and not sonnet 5.5.
>>109932119I saw responses as low as 18ms on my 7900xt. I tested 512 and 1024 token context. It can process more than 600 requests per second here.
is sonnet the terra of claude
>>109932222it was supposed to be, but even luna mogged it.
>>109932135Maybe another context window reduction will do it? 192k for the next model perhaps?
>>109932135
Gemini 4 is going to be like nothing we have ever seen before.
>>109932246Gemini 4 is going to be like nothing, like we have seen before.
>>109932246pooperintelligence
>>109932135devday tomorrow
>>109932246Gemini 4? Best I can do is Gemini 3.9 Flash-lite V12.395.2026but don't worry, we just began training Gemini 5 Pro
>>109932242
Anthropic models are thoughtcels now.
>>109932261>"we have the compute, just turn up the thinking!"kek safetyists really want to lock into this architecture
>>109932261Max is just short for benchmax.xhigh and high looks fine.
>>109932261max is 99% not worth it
>>109932261It shits out so much tokens it's more expensive than Opus in AA.
>>109932261chinks would kill to figure out what is going on inside Anthropic's CoT
>>109932090
>>109929866is this good enough? it shouldn't be difficult for top of the line model to port a shitty old unfinished version of minecraft to c++ . the code is right in front of it
>>109932307i just talk to my clanker like "boy this shit still ain't fixed after 3 tries, pick cotton"
>>109932261use case for thinking?while mogged by luna, at least terra still set itself apart from solwhat's the use case for this tokenmaxxing, it is not deepseek flash
>>109932317>use case for thinking?mogging Astra?
>>109932317>use case for thinking?I don't think there are any, I stopped doing it irl and I'm fine
>>109932316also "chop chop clanker boy [rival model] is wrapping it up already"
Why did nobody mention that Docker Sandboxes is closed-source?
How large is the Sonnet model?
How to vibegod a big black cock in my ass t. indian
Can Sonnet 5.5 orchestrate well?
Hey, bastard who told me not to use my reset... where's the reset today...
>>109932355Sonnet 5.5 (medium) orchestrating Luna 6 (Max) is probably the greatest poorfag combo ever
>>109932355The only use case for Sonnet is speed. Or if you for some reason need a very recent knowledge cutoff date.
>>109932374the modern tokenmaxxing
>>109932374I waa hoping to use it as a router for Opus 5.5/Astra/Grok 4.7
>>109932374sonnet 5.5 medium orchestrating v4.1 flash agents?
>>109932374>orchestrating Luna 6 (Max)mogged by deepsneedluna might be cheap but it's not a good model
Honestly the strategy is genius, only release the short stacks, the Chinese can't distill from your top model, keep the frontier comforts for yourself. Easy game.
deepseek will probably invent continual learning next
China in shambles
yeah I'll use Sonnet 5.5 with Deepsneed for work then save Opus for my personal project, kek.
>>109932396??? What do you mean? The short stacks?
>>109930028we're a deep consumerist society beyond repair at this point. normal cattle don't have any imagination since the 70's when mass consumer manufacturing came into existence
>>109932386If you do per turn routing you just fuck with your cache and you will pay more than just using the larger model.If you want to do real orchestration the better thing is to use a smarter model and just tell it what it should do, what model to use under what circumstance.
>b-but muh devday!KWABCOPEX
>>109932396I want a short stack for myself.
Xfags claim that Gemini 4 Pro just leaked and is gonna smoke Openus 5.5
>>109932495how much you want to bet that this originated from an indian account? or pakistani?
>>109932495>if true
>>109932495It's just some insufferable faggot trying to engagement bait you.
>>109932429
>>109929968kino if you can completely dismantle and destroy a planet for its resources
>>109932495Thanks; you fed Rajeesh for the next month
>>109932520why post a dead pdf game
>>109932537ask your llm
>>109932540it said that you were most likely trannoid
Pelican status?
>>109932545based. I'm glad it likes me
>got a x.com account because there are so many links in this general>my timeline is now full of vibed projectspretty fun to be honest
>>10993256610 trans folx just died from you supporting musk facist
>>109932189Yeah in my cluster of H200 nodes that I have in the yard
>>109932189@gork what's the street price of 1.5 TB ddr5 ram? At least 6000 mhz, to be used with a modern epyc cpu for octa channel ($15k btw)
>>109932566The following tab loads only a handful of tweets nowadays. Can't even curate shit. Complete shit.
>>109932566Think of the water you're wasting, chud
>>109932628thank you for the free hardon
>>109932566I had a x then they wanted me to use my phone for verification.
>>109932628I wish I was taking that water from starving africans, but I'll settle for taking it from crying farmers
Yeah I think I'm cancelling my Codex sub
>>109932660Son, me and your mother will get a divorce.
>>109932660I'm waiting for the next frontier release. Don't screw it up Tibro.
i just test gpt6 on arena to fix userscript. it sucks. gpt5 way better than 6.
My main issue with Sonnet is that the cost per task is still super high on the low and high ends and competing OpenAI models are cheaper and comparable in intelligence. I can see a case for medium and high possibly if you were Claude Code exclusive but honestly on a cost basis, I don't know if it is even that much better here.>>109932660I'm waiting on dev day but I am possibly cancelling as well. Opus 5.5 is just too good.
>>109930293>>109930312>>109930330>>109930377>>109930385Thanks
>>109932660I have my $200 plan I can't get back if I stop so I will stay for now, there is always leapfrogging going on, openai will be better again next week
>>109932714If they bring the price of Astra down by half I will stick around. Otherwise I'm leaving to Anthropic, my sub is already set to cancel on the 6th.
>>109932660Everyone should do this so i get better value for my sub.
Opus 5.5s outputs are so CLEAN5.5 medium is so FAST
>>109932765
I had my first user download omg omg omg
>>109932753This is why you never buy the yearly plan or take the console war stuff seriously
>>109932771I swear it's true. Clean might be a little more subjective but it's stupidly fast for what it can do.
>>109932782>This is why you never buy the yearly planI stopped buying yearly plans after I got ultra scammed by HBO MAX on my $250/yr plan. They got rid of 4K streaming, and I had to use a dogshit version for 6 months.
>>109932797It's definitely faster than Opus 5, considerably so. Depending on the task it can still take a while, it's not like a model running on cerebras
I hate how well this claude code fucking woke jew shit handled my first coding prompt.
>>109932817I'm only talking about medium. On max it's slower than fable.
Not gonna use claude even if it's better than codex. Anthropic can't be trusted for shit.
>>109932863>OAI can be trusted
>>109932863>i'm not stabbing myself in the cock even if it's better than lighting my nutsack on fire great job goy
>>109932863Why?
>>109932875>>109932877>>109932879There is a difference between getting a bad deal and getting stabbed in the back because Dario had a bad episode and decided to fuck everyone over supporting the [current bad thing].I shouldn't even have to spell it out, you're crazy to use anything from them.
>>1099329132 snakes = 2 snakesthank you sundar pichai
>>109932918Yeah you're retarded I got it the first time
>>109932925if you don't want a response don't click the post button lil babby
>>109932863In my view, giving Anthropic money increases the chances of a good IPO because it improves their financials. Which in turn increases the chances of regulatory capture of the whole AI domain because the entities Anthropic is commingled with (like METR and CAIS) may get the powers they're after. Those people, and by extension Antrhropic, may get to decide the future of things like open models and model training in general.
>>109932949Democrats will be in power by then, and they do not allow this kind of corrupt corporate nonsense.
>>109932913>OAI would NEVER stab me in the back, not even a little bitThe only difference between these companies is the quality of the product they offer. The PR stuff from both are equally performative.
>>109932949They clearly don't want open models to stay around nor people competing with them, so why even support them
>>109932955>muh party linesPeak snailcat mentality.
>>109932955kek what, you're on the wrong website
>>109932956>literal schizo at the top that even scares investors>it's the same thing brolol
>>109932977Is Sam more likely to team up with Dario, or you? Consider carefully before answering.
>or you? Consider carefully before answering.
>>109932981Their interests are lining up for some things or they'd have merged already
>>109932977Stop glazing Dario.
>>109932875OAI can be trusted to fleece people for their money. Anthropic is a messianic cult.
109932918>troll posts>gets called out>seetheslol, lmao. >>109932932 is right. dont click post little baby. back to blusky you go.
>>109932995>they'd have merged alreadythat is not how this system works at all were you born yesterday?
>The company they tell you to not worry abouthttps://archive.is/QMo1k
>>109932995>or they'd have merged alreadyHart-Scott-Rodino Act and Section 7 of the Clayton Act
The chinese creating "thinking", they're basically machine elves.
>>109933037
>>109933037getting sucked into a machine elves LMAO
>>109932955Don't look up the political views of early Ant investors who are also AI safety donors.
>>109932482Okay, what I have going right now is a session of Opus that I communicate with. They then communicate with different orchestration sessions, which then delegate tasks to sub agents. Ones for auditing and two are for planning different aspects of project/delegation.
>>109932495World better prepare its anus for Gemini
>>109932997>OAI can be trusted to fleece people for their moneyAnd steal your shit if you make the mistake of stumbling on to something significant
is the Codex model degradation thingy true? is it time to switch to Claude? my coworker has been mogging be since Opus 5.5 released
>>109933075Hey chud have you ever heard of effective altruism?
>>109932918badger badger badger badger
Sol is so bad compared to 5.5 I want to cry but I feel like if I'm paying for this I should use it
>>109932875>OAI can be trustedImagine if this worked and if Anthropic had a convenient opportunity to absorb OpenAI.
>>109932689>My main issue with Sonnet is that the cost per task is still super high on the low and high ends and competing OpenAI models are cheaper and comparable in intelligence.??AA index is for general intelligence not coding. sonnet-5.5 is a continuation of opus-5.5 at the lower end. so again get beaten by on price by gpt-6-astra and gpt-6-sol.for coding however, opus/sonnet-5.5 use less tokens and are cheaper. astra/sol can't compete.
>>109933167>if Anthropic had a convenient opportunity to absorb OpenAI.regulatory impossibility
I do inauthentic arguments with myself about openai vs anthropic to keep them trying to invest more to please me
>>109933170De facto it could have gotten most employees.
109933108poh ho ho someone is triggered. notice how quick he runs to the word "chud" to frame things in his favor and try to put the other poster in a "negative" frame. how festive. notice how he lies and implies its "altruism". what a good little cock sucker for whatever side he leans. comical how triggered he got out of nowhere over a word or words. lol, lmao.
Harvard did an RCT on class room learning at Harvard vs. with ChatGPT (4 at the time) on undergrad physics. Turns out that students learning with AI outperformed the class room learners by around 30%. They were also more engaged and more motivated and spend less time than they normally do in class: https://www.nature.com/articles/s41598-025-97652-6>tfw you can literally vibecode better teachers than harvard (which costs around $100k/year) on a $20 sub
>>109933197we shippin' headcanons now
>>109933197only fanatics like dario just give up their openai stock
KEK /ourchud/ did an ebin memecopex tards in shambleshttps://x.com/TimSweeneyEpic/status/2104653127618089439?s=20
New thread:>>109933222>>109933222>>109933222
>>109933205literally shatbpd fucking 4 on PHYSICS
>>109933212quite a few safety people went from OpenAI to Anthropic, like Jan Leike who's head of alignment (those are more likely to be fanatics apparently).
>>109933205vibeGODS stay GODDING
What Is Google Cooking?!
>>109932086>It's crazy how garbage Sonnet 5 was too, at Max effort it still gets destroyed by 5.5 lowsonnet 5 was more usable than opus 5 at orchestration for me while not blowing through weekly usage.
