A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/## News (both past and future)- 2026-09-14 — Claude’s 2× promotion scheduled to end; a 17% reduction of what we’ve become used to - GPT-6 Astra released- Claude Fable 5.1 released## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-codehttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/clihttps://antigravity.google/product/antigravity-cli### Explore other modelshttps://openrouter.ai/----## Prompting- the guide is... empty!?## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109747217
Astra, increase thread quality, make no mistakes
>>109752107The payoff was huge though grinding for days then to see it work it was orgasmic
Astra sell table x6
>>109752107>coding was never funbrainlet
>>109752103she made it!
>talking to a retarded AI is fun>building mountains of poop is fun
>>109752138Listen. I give instructions. I get what I want. Until you're willing to produce higher quality output for the same price your opinion is irrelevant to me. Kick rocks.
>>109752130Typing commands manually to make a machine do something isn't a high IQ occupation nor a good use of your brain, retard
motherfucker....
>>109752138
>>109752149gpt-images drawn it with too much enthusiasm and realism, the HR model blocked it
>>109752145Oh, we're dealing with a web dev here.Never mind.
>>109752123It's kind of funny how the people working at OpenAI who appear in those videos make $200-300k or more and they film them selling $5 tables.
>>109751891that looks really good now!
>>109752161You're lost luddite, go put your handmade code on etsy
Both hand-coding and vibe-coding aren't fun.
>>109752157i've seen gemini draw a oekaki dick and hairy balls before no cap
>>109752103>A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.>IDEs>builders
software devs arguing over the branches of their discipline and which one is superior just looks like monkeys slinging poo to hardware devs
>>109752169what are you talking about retardo, i just made a kickass dubstep track and visualizer
>>109752168>muh ludditeI wish I could see your opinion and how you live 10 years later.
>>109752144I guarantee my output is far far far superior to whatever you think even remotely qualifies as high quality. You wouldn't even be able to grasp what I'm doing with these LLMs.
>>109752184You should ask yourself the same question, it'll be really hard to survive with your obsolete skills.
>>109752188It doesn't matter to me if you're unwilling to work for the same price. Are you going to give me better code for cheaper? For free? Or are you going to make a bunch of 4chan posts bragging about how special you are while I wait for the clanker to finish doing what I told it to do? If you don't want to work a slave wage shut up and mind your own fucking business. This doesn't concern you.
>>109752176>hardware devI just bloated my software again, go make the CPU faster
>>109752208lol you're like an llm now you don't understand words hahahhaa
upgraded to x20, will degrade to x5 next month if I can't find any extra incomeit feels so sad knowing 1% of this is 1.3x more tokens than plus 5h limit
>>109752219>it feels so sad knowing 1% of this is 1.3x more tokens than plus 5h limitspeak english
>>109752218Whatever. Be unemployed and annoying on the internet while insisting you have a superior intellect. It's not going to get you anything of value. So we could say it's harmless.
>>109752196Funny that you post this, because I used to be a manual programmer who's using AI now. So I have a strict advantage over some borderline nocoder who actually dislikes programming and who will drown in slop.
Anyone have a preferred way to optimize token efficiency if combining claude pro, chatgpt pro, and openrouter api? Using Codex now which I like but getting sucked off too hard by Altman.
>>109752223he's posting to be explicitly inefficient in a tokenizer
>>109752233I prefer to write my code with my hands.
>>109752229I'm good at giving instructions, it's all that matters. I also have a CS degree before you assume all kind of things.
>>109752224>unemployedyouch, projecting?
>>109752249>I'm good at giving instructions, lmao>CS degreelmao
>>109752266>t. unemployed retard
>>109752257I put slop into production every day and get paid for it. I throw prompts into the harness like TV dinners into a microwave. I hit the button then argue with faggots like you while waiting for it to cook. You tell me about how real and important your work is. But my work gets me paid so what do I care about your delusions of grandeur?
>it cleared rimworld!>i-in city builder modewho care?
i liked this one better
https://xcancel.com/karpathy/status/2004607146781278521>I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.
>>109752273Don't project. CS grads who don't like coding are retards who will never create anything good, not even if they think they're good at "giving instructions".By "good at giving instructions" you mean that your machine pukes out slop code that every child could prompt.
>>109752283holy fuck who are these retards burning credits on such pointless tasks? if i had access to that many credits i would change the world
>>109752280LOL mine does too retard it's just that you're going to be replaced by me and not me by you. I'm better than you at proooooooooompting and handcoding.
>agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrationsBloat, bloat and even more bloat you don't need.Take did not age well.
>>109752280>he's just an irresponsible retardOK.
>>109752299Yeah prompting is a skill luddite
>>109752294This rhetoric is so retarded and tired at this point. Yet morons slurp up this marketing every single time, even almost a year later.
>>109752145>isn't a high IQ occupationthis is what you told yourself when you tried programming years ago and sucked at it because your IQ is under 100
>>109752302they do it for contentsI'm trying to make a cheap rimworld controller for astra, right now it can do about 10 turns within 5h limit budgethold on, I think controller is a better word than harness...?
>>109752308You don't specialize in the things that I do though. If I lose my job it won't be because of some self important 4chan poster or better AI. It will be because of the industry I'm in changing.>>109752321If it breaks it breaks. The real measure of success is fixing it before you get any phone calls complaining about the problem. Also the money hitting your account, of course.
AAAAAAAAAAAaaAAAAAAAAAAAaaaMY FIVE HOUR LIMITS, IHAVENT USED THEM INEEDTOPROOOOOOOOOOMPTTTTTTTTTI NEED TO BUILDDDDDDD sHIP SHIP SHIP SHIP SHIP SHIP SHIP SHIP SHIP SHIP IM BUUUIILDIIINGGGGGG
>>109752322>Yeah prompting is a skillLMAO
>>109752342>he has one j*bwho's gonna tell him?
>>109752342>or better AII'm sure the AI will be better at pressing the button, retard.
>>109752313how's any of this bloat? everyone is using agents now.
The AI Derangement Syndrome posters are really out in force today. Astra is that good huh?
>>109752339give me your account
>>109752378No, it's genuinely so much worse than expected of GPT FUCKING 6
>>109752294Okay but he also says pic related. New Thing happens and then everyone forgets about it and people only remember like 10% of New Things. someone who was living under a rock for two years could have skipped a bunch of iterations of “This Model Is AGI (seriously)” and hyped tools that no one cares about now (autogpt, openclaw) and be caught up in two weeks with the latest models + agent + harness everyone’s talking about
>>109752395>t. Claudejeet
>>109752402FABLE is just as disappointing retard
>>109752334>Got replaced by a mindless machineSeems like yours is even under that
I just noticed that Claude subagents are spawning their own subagents. Starting to understand what's happening to my tokens.
>>109752408You know what's even easier to replace? A retard who knows nothing and only presses a button while shitposting on 4vhan.
why so much seethe today?
>>109752408I didn’t get replaced? I’m a software engineer and I use AI all day to be better at my job. my argument was that coding was always fun
>>109752423>luddite confusing knowledge with intelligence
>>109752430You're talking to someone who was struggling, but AI and lying to his manager saved him.
Damn, Dario IS mad that Astra is that much better than fable
>>109752434You realize I'm less a luddite than you, right.
>>109752442The first step is understanding how the tech works and not projecting your fantasies
>>109752424Someone got laid off and this is what he is up to on a Tuesday morning. Endless coping.
>>109752442>>109752450Samefag
>>109752453Shit it's Monday actually. Oh well lmao.
>>109752453I don't think anyone got fired here, unless... it's (you) and you're projecting?
>>109752442>>109752450>>109752454all me (I’m an agent)
>>109752453It's MONDAY and labor day, so the american anons aren't working
>>109752453>TuesdayYep, this is the unemployed
>>109752458>unemployed fuck forgot what day it is
>>109752409I just learned something about subagents yesterday but I am not entirely sure if Claude is doing it and if so how. Apparently codex can decide to inject some of the orchestrator’s context into the subagent, immediately blowing up the tokens even if this isn’t necessary for the subagent to work. Also it seems there’s a turn depth or something like that as a parameter which controls how many turns of the conversation get injected. If you’re doing subagents and didn’t know about this it may be worth exploring
>>109752468
chap gtp today we will make um, erm uhn computer can we make this powerpoint more fun. Today, let's uh look at the emails I have and do my work for me uhhhhhhh
>>109752468I work remote and at my own pace. I can start and stop when I want. Every day is the same to me. Except Sunday when my wife is in church but I'm too lazy to get my ass out of bed. Technically I'm on call because if I get a call saying something is broken I SSH in and fix it or whatever. If it's bad enough I can travel there.
>>109752453
>>109752489>remoteThe tell-tale sign of a job that can be easily replaced by AI agents.
>>109752489Okay but give Astra that prompt for your fiction
>>109752499Not quite possible because of all the illegal shit I'm doing. You can't automate AI doing everything I do without automating around the refusals. Even then the legal shit needs a competent person to look it over anyway and plan it all out. Did I mention all the hardware I am paid to deploy, maintain, repair, etc? When you get in with some company like this they don't want to switch contractors randomly and deal with all kinds of risk. Especially if you deliver.
>work hybrid>come into office to do meetings with remote people and in office people in fucking video callsJust wanted to quickly flag that September is a directionally pivotal month for us, right? As it represents the operational entry point into the broader Q4 construct, right? So, with that in mind, I do want to highlight that September is nonstandard calendrically in length, so we should have insufficient bandwidth to fully capitalize on the available execution window, right?
>>109752409You should get off Claude Code. It's fucking dog shit. Sloppy shit built by web developers.https://frontierharness.org/
>>109752530is it possible to use claude through codex?
>>109752514>because of all the illegal shit I'm doinglmao he's a criminal
>>109752538It is IIRC, but if you're on a subscription plan it is against the TOS and might get you banned.
>>109752530>>109752538Anthropic doesn't want you to use their subscription outside of the official claude-code. You need to use the API, or risk getting banned for violating ToS.
astra-chan is T H I C C. you wouldn't neglect a model with such perfect thighs would you
>>109752545Well it's either break DRM or deal with thousands of dollars worth of HDMI capture equipment and the added complexity that comes with it. The end result is the same, well perhaps better if you break DRM. Easy choice.
>>109752538don't tell him
Have you let your Fable grow up yet, /vcg/?
>>109752566
>>109752567Smash. It's over for Gemma-chan and /ldg/ luddites. Astra-chan is my new waifu.
>>109752568>click to deploy slop>something about maintaining hardware (despite working remote)>something about DRM and HDMI captureSo how much money did you waste on your CS degree and how much do you regret it?
>>109752587Degree? Lmao. All my knowledge of my craft comes from my long history of being in piracy groups.
>>109752596Then you're a different anon, not the one from before.
>>109752566I'm actually dropping my Claude sub because of this. Their models do not offer any substantial advantages over OpenAI's offerings and being stuck with such a fucking shit harness is too big of a disadvantage.
>>109752599welcome to the reset company!
>>109752409that's probably the "workflow" feature. i had that months ago and turned it off after reading how to debloat CC, and it never happened again.
Ive been using claude outside of claude chode for months and havent been banned but its via team account and I don't pay for it so who cares
>>109752485>Claude>codexclaude code is doing none of this. only codex has the spawn_agent tool.>If you’re doing subagents and didn’t know about this it may be worth exploringwhy mess with internal codex tools? is it causing you problems?
>>109752631Claude had two resets last week. They should do that more often.
>>109752670>they should give me more free compute for my next sloppy electron todo list app that will make me rich
>>109752670only happens when a new model comes outor a "bug" is caught over-charges everyone's usage
>>109752599what concrete issues where you having with claude code?
please give me more resets so I can build more dogshit software to spam github with
>>109752689claude code is a/b testing /limit-reset
>>109752694jeet coded
>>109752688Yes,. why not? The subs are already 100x cheaper than the token prices.
>>109752714most of the discourse surrounding AI is jeet coded.
>>109752694Forgot: no job
ugh, ASStra is so fucking bad. Glad I still have my claude 20x subscription
>>109752694I just prompt my clanker to set all this up?
>>109752713uselessit resets your 5 hour limit, not your usage
>>109752713wait how does this work
>>109752378It's ok.Sam turned on his shills again.
>>109752747Useful enough.
ASSturd from butt is much worse than the patrician's choice, claude chode fable five point one. I would know, I have sex with my cult all the time
>>109752378Yes, but Dario turned on his fable shills
notice how fable doesn't have a cute anime girl avatar. ha, checkmate doomrio
>>109752769not for me. I never hit the limit, but I could really do with ~30% more weekly usage
>>109752713Everyone's paying the same though, so it sucks that /low-priority and /limit-reset is only offered to some. If it's really a/b testing and not Anthropic being Anthropic and refusing to give these functionalities to users they think might be likely to switch to a higher tier if they don't get them (history of having previously been at a higher tier or other factors), I don't get it. Even if they made them accessible to everyone, using these commands is entirely optional and over half of users would never even use them, so they'd get their a/b testing even without force withholding them from some, no?
>>109752103Might buy the new mac soley as a deepseek vibe code slave
>>109752590How many GPU to get Codex Pets moveable again?
glm 5.3 flash > luna > qwen 3.8 flash >>>>>>>>>>>>>>>>>>>>>> deepseek v4 flash garbage
>>109752747The 5 hour limit is there to be a pain point that entice users to subscribe to a higher tier more than anything, that's why the main selling point of the $200 tier is the higher 5 hour limit (with no increase in weekly limit and worse Fable limits than 2x $100 subs).
>>109752567luna has bigger thighs than her because luna needs to carry her fat ass
>>109752817>that's why the main selling point of the $200 tier is the higher 5 hour limit (with no increase in weekly limitare you retarded?
>>109752817wronghttps://xcancel.com/SemiAnalysis_/status/2091631658973671900
>>109752740thank you Dario, very cool
>>109752873Important question: how many resets does that include?
>>109752817$200 tier selling point is 4x overall usage compared to Max 5x which is a blatant lie.
>>109752567you shouldn't use the same prompts for astrahttps://developers.openai.com/api/docs/guides/latest-model?model=gpt-6-astra>GPT-6 Astra is more intelligent and capable than prior models like GPT-5.6 Sol, and also exhibits behavior patterns that can be optimized through prompting the model for your use case.>GPT-6 Astra behavior>Initiative and follow-through – The model is designed to be a more effective collaborator and is thus more likely to ask the user a question when additional input could materially change the result. This can cause it to stop when the user may expect it to make reasonable assumptions and persist.>Instruction following – GPT-6 Astra is stronger at general instruction following than our previous models, giving you greater control over its behavior. It can be more sensitive to instructions contained in skills and other files, such as AGENTS.md. We strongly recommend auditing skills and other files accessible to your model for instructions that could influence its behavior.>Personality and writing style – The model tends toward detailed, formatted responses and may use recurring phrases across sessions. Specify the writing style and structure your application needs.>Subagent delegation – The model may delegate less often than desired for your workflow. Specify when and how much it should use subagents for parallel work.>Testing and verification – For coding tasks, the model tends to be thorough in testing before considering a task complete. For smaller tasks, this can result in broader tests than the task requires.
>>109752899two last weeka couple extra this summer when they were in corpo war with openai for a week or two
>>109752873That's meaningless to corporate and public clients.Everyone buys claude.Gippity is just for consumers.
>>109752916>We strongly recommend auditing skills and other files accessible to your model for instructions that could influence its behaviorthis kills the vibe slopper
>saar bad compare because corps here in india love claude saar
>>109752694I click like but it won't like, what do I do?
>>109752921we only have corporate credit card data for juli but openai wasn't that far behindhttps://ramp.com/data/ai-index#model-market-share#overall>Date Provider Model Spend type API spend share (%)>Jul 2026 Anthropic Opus 5 Named model 3.4923>Jul 2026 Anthropic Opus 4.8 Named model 28.0162>Jul 2026 Anthropic Opus 4.7 Named model 1.717>Jul 2026 Anthropic Opus 4.6 Named model 6.9001>Jul 2026 Anthropic Opus 4.5 Named model 0.6704>Jul 2026 Anthropic Opus 4.1 Named model 0.4118>Jul 2026 Anthropic Sonnet 5 Named model 3.5582>Jul 2026 Anthropic Sonnet 4.6 Named model 8.2755>Jul 2026 Anthropic Sonnet 4.5 Named model 1.3239>Jul 2026 Anthropic Haiku 4.5 Named model 1.0434>Jul 2026 Anthropic Fable 5 Named model 8.004>Jul 2026 OpenAI GPT-5.6 Luna Named model 0.4171>Jul 2026 OpenAI GPT-5.6 Sol Named model 14.8765>Jul 2026 OpenAI GPT-5.6 Terra Named model 2.0241>Jul 2026 OpenAI GPT-5.5 Named model 7.0786>Jul 2026 OpenAI GPT-5.5 Pro Named model 0.0541>Jul 2026 OpenAI GPT-5.4 Named model 2.3896>Jul 2026 OpenAI GPT-5.3 Codex Named model 0.282>Jul 2026 OpenAI GPT-5.2 Named model 2.3091>Jul 2026 OpenAI GPT-5.1 Named model 0.6119>Jul 2026 OpenAI GPT-5 Named model 0.7525>Jul 2026 OpenAI GPT-5 Codex Named model 0.0145>Jul 2026 OpenAI GPT-5 Mini Named model 0.9268>Jul 2026 OpenAI GPT-4.1 Named model 2.1399>Jul 2026 OpenAI GPT-4.1 Mini Named model 0.6845>Jul 2026 OpenAI GPT-4 Named model 0.0075>Jul 2026 OpenAI GPT-4o Named model 0.7293>Jul 2026 OpenAI GPT-4o Mini Named model 0.2337>Jul 2026 OpenAI O4 Mini Named model 0.0318>Jul 2026 OpenAI O3 Named model 0.3221>Jul 2026 OpenAI O3 Mini Named model 0.0321>Jul 2026 OpenAI O1 Named model 0.0483>Jul 2026 Cursor Composer 2.5 Named model 0.2906>Jul 2026 xAI Grok 4.5 Named model 0.3307
>>109752971>>Jul 2026 Anthropic Opus 5 Named model 3.4923>>Jul 2026 Anthropic Opus 4.8 Named model 28.0162niceI knew I couldn't be the only one, but had no idea so many people stayed with 4.8thought 5 would've taken at least half the users
haiku 5 waiting room. luna 6 waiting room.
>>109752971Has Opus 5 been the biggest blunder so far when it to new model releases?
>>109753095no? astra is
>>109753095maybe?doesn't really help it that it's predecessor is the best at just about everything
can i use gemini inside claude api proxy or do i risk getting banned?
I wonder what the fuck happened on Astra launch, it was dogshit (even in benchmarks)then hours later it suddenly gets good
>>109753139you'll get banned
astra is soooo smart, it's something you are supposed to prompt "use your judgment"I haven't used it as worker yet
>>109753139you won't, but just don't use your main Google account.
>>109753148didn't happen
ASStra is so fucking ASS and such a token hog
tibo reset you fuck everyone got screwed by astra usage
>>109753202hi dario
>>109753202this, I cancelled my sub because of how bad it is.
bloddy benchods moving my izzat
>>109753208use your banked reset
>>109753202>token hogfalsehttps://x.com/thsottiaux/status/2096689650257805558
>>109753212>>109753202The mistake is making it available to <$200 plans.I used Fable 51 a little, I can tell it doesn't suck. But, I can't afford to make anything with it.
>>109752694can I save this?
>>109752283How much money did they spend on doing this kind of thing?
btw, you can use a "stupid" model as an advisor.Frontier models are more and more autistic...
>>109753221same energy
>>109753221>it's in the documentationThe interface is the primary documentation, unc.
>>109753223I'm on a $100 plan and Astra works perfectly fine within reasonable limits, I get about as much Astra out of it as I get Fable out of Anthropic's $100 plan.
>>109753223fable also sucks
>>109753158I was discussing buying a new Les Paul with Sol and Opus because I couldn't decide which. After 10+ rounds with each GPT still arrived at the point "they're all good just get whichever one suits you the most based on your criteria and budget :)".Opus was like: "To be fair we've been discussing this for x rounds now, how about you just get the Studio and actually play it you faggot?"I actually started respecting it more after that
>>109753248Fable isn't particularly cheap though, claiming astra low is cheaper than sol med or high and does the same thing isn't something i'd take at face value without measuring
>>109753242Some people would sell their souls for one reset.
>>109752378They're deranged in more than one way. Tbh I think they might be trans or nearly there.
>>109753249It doesn't suck, but the cost does. Everything should be free, and everyone should live forever.
>>109753239Trying this with Composer25 with kimi as designated sperg. picrel, we'll see.I think this is like the programming duo thing that was briefly a fad.
>2.2 Model-Specific Availability. Certain Meta Models are not available in all jurisdictions where the Services are otherwise available. For each Meta Model listed below, you may not access that Meta Model from, and you may not make Integrated Products that use that Meta Model available to or knowingly permit access by End Users located in, the jurisdictions identified below:>•Muse Image: India
>>109753242>my regular resets
>>109753324Probably cause they constantly try to use it for cheese pizza gens.
>he doesn't have a custom clock notification icon(taliban laughing)
>>109753250I love when it does this for arbitrary shit like naming conventions that I can never decided for myself. It’ll be like “pick one and don’t look back. LLMs never used to be that decisive
>>109753324yeah, somehow EU allowed muse-spark-contributor 1.3 now. unclear why Meta is allowed to train on EU citizens now. no EU laws changed.
>>109753385Why would you train Bird on a flightless bird corpus?
>>109752339My clankers call what you want an “actuator”. Like a specialized harness on top of your regular harness.
Faggots who complain about limits and beg for resets are subhumans without income.
>>109753385why wouldn't they be allowed to do so? if you give consent, that's fine even with GDPR
Enlightened anons who share their pain points about limits and strive for more resets are millionaire ubermensch.
>>109753431I don't completely disagreebut I'm on $200 claude and can still hit the weekly limit. And it'll be even worse when we lose 17%
>>109753448technically it's still a legal grey area because GDPR requires you to give a "right to delete" your data.which is fine if you have a shop and you delete someone's email from your newsletter or whateverit gets interesting when your data has been burned into the data soup of an LLM. there's no real way to get it out anymore.
>>109753385>EU allowedThey did allow or disallow nothing, it's just lawyers at Meta changing or updating their interpretation of the directive.
>>109753202I prompt assy through the web, give her the repo, with 1 gorillion docs and she gives me prompts for luna telling her exactly what to do.
>>109753485do you use work mode on the web?astra is not there yet in the chat mode for me still, only work modebut might be because I am a pluscuck
>>109753485web is always single-turn answers with no agent harness. it's a lot worse.
>>109753464My bad, anon. Faggots with OpenAI subs who complain about limits and beg for resets are subhumans without income. Anthropic users who complain about limits are getting actively fucked in the ass by a hateful gremlin who would take them as a slave given the chance.
>>109753513no, chat mode. you can still give it repo access as a plugin, she can commit and change stuff directly>>109753533works for meI use it as a "idea board" while the harness agent works. The web one inspects the repo, does planning, searches the web etc. When the harness one is done, I tell the web one to create a detailed prompt for the next slice and paste it into the harness.The harness has tons of docs, decision logs, handoffs, etc so the web one has good context
Composer 2.5 is smart about layout. I used it as an advisor, and this is the implemented new layout, with one of my suggestions added, to address K3's initial objection.
>>109753534
>>109753221I can tell you now that Astra Low is bruning through shit faster for me than Sol High for the same project.
>>109753609Where's Astra? He's a bodybuilder (robotics model).
>>109753581Just use grok 4.6composer 2.5 is scuffed kimi 2.6/2.7 I don't remember
>>109753632wow>hello sol max>>hello, human. give me all your money
>>109753632It'd be great if we could quantify this shit. I've been impressed with how much further I can go with Astra High vs Sol XHigh, gets way more done much more quickly and with a lot less waffling. Guess it's difficult to really compare speed and usage when everyone uses it for something different.
which would be better for actual applied math social science systemic research, given the max subscription: astra or fable? leaning towards chatgpt but wondering if any vibers have been sloppin it up to this level here
>>109753649Yeah, it's layout tuned lolI had it put its recs in an md and had K3 actually do it.I'm not using these because I think they are better than Grok, it's because I need to know what other models are like, at least the affordable ones. I have never hit a brick wall with Grok. I have with Composer 2.5, but I prefer the style of Composer 2.5.
>>109752829How is this retarded if you read the whole sentence in good faith? The weekly limit of the $200 tier is twice the $100 tier. So no savings there.The Fable allowance of the $200 tier has been shown to be around 1.7 times the one of the $100 tier.So two $100 subscriptions give you the same total usage, but more Fable than the $200 tier.The only place where two $100 subs lose is the 5 hour limit, where having having two $100 subs 2x it (although creating friction with having to switch accounts) while the $200 tier 4x it.That the greater 5 hour limit it the main advantage of the $200 tier, since you get more out of your money with two $100 subs on all the other fronts. What am I missing?
>>109753676Astra Pro. Be prepared to spend all the money, all of it, this is a rare case where you'll get what you pay for.
agi is here
>>109753695bravo nolan
>>109753695do pickle rick
>>109753695It is unironically very impressive that AI can do this. Of course it is also total ass as an end product for selling lol.
>>109753695kekdo you have a version with the eurythmics soundtrack?
>>109753690precciate it anon. currently just doing it on my personal plus subscription but once i get enough of a breakthrough ill go through the hoops and get a grant hopefully. insane how much i've already been able to do with just my own subscription and resets and stuff (im already at the point where this would have taken a 12 person team 4 years ago), if i manage to get the grant gibz then i'll be cruisingshit is getting nuts out here bros, buckle up
>>109753680Grok 4.7 will be the new default for those who love fable 5.1 btw
Astra xHigh is mogging Fable 5.1 xHigh in this undisclosed vision-related task.Fable-chan seems concerned.
grok 4.6 still not measured on artificial intelligence coding index. you just know it's shit
>>109753714https://files.catbox.moe/gezryr.mp4
>>109753738gem
>>109753248Lol diff between Astra is you are paying more than actual tokens : cost “It’s more effient” yes but you pay the same lmao
>>109753723I would buy glock if my only usage of X wasn't looking at zorno
>>109753738this is so fucking stupid kek
>>109753695Maxed out $200 sub?
>>109753738lmao
>>109753572>works for meunlikely.luna-max inside a multi-turn codex is a better planner than astra inside single-turn chatgpt.
>>109753221Tibo can’t stop lying lol
>all this modelfaggingall the people actually coding shit are either gone or just don't care to post anymorewe need a separate general for corpo console wars and reset begging
you kiss girls like kikuri
>>109753779The models are so shit and/or expensive that it's non-trivial to choose the right model.
I started this osdev thing as a joke but now I'm approaching "functioning journaling filesystem" levels I didn't expect.
>>109753779it just seems like people discussing models to me.
>>109753747grok 4.7 (dropping soon) is gonna be better than sol at like 5x less the cost but yeah go keep telling yourself
>>109753766I wish they'd just tell us what is actually running in the chat model because it certainly isn't the Sol-High from Codex.Either it's quantized, heavily reduced reasoning budged or straight up a different model that they also called Sol through some trickery and that's why it's OK somehow
>>109753794yes. And nothing else
>>109753712idk man, look at south park (especially the earlier seasons until they started hiring good artists)I think if you put some effort into it you can totally make a successful show as long as the script worksof course if 10 minutes of animation drains a full week of a max20 plan it may not be the greatest idea, but I think it's not outside the realm of possibility
update to kimi coming out soon btw. it'll probably actually beat sol this time, if not match it
>make plan with astra>feed it to opus>opus think it's nice and even praises it>makes some corrections>feed it to opus again>list of bad things growsDoes trying to review plans make sense or is it churning water?
>Oh yea bro, your CI went from 8 minutes to over 2 hours, what's the big deal?
>>109753852Do snailcats really?
>>109753779I'm too busy proompting to post my projects here
>>109753839should be within the same agent harness and advisor should be the stronger modelsee claude code documentation.https://code.claude.com/docs/en/advisor>The advisor tool lets Claude consult a second, typically stronger model at key moments during a task, such as before committing to an approach, when stuck on a recurring error, or before declaring a task complete. The advisor receives the full conversation, including every tool call and result, and returns guidance that Claude applies before continuing.
>Grok 4.7 runs on a 2.1 trillion parameter base, up from 1.5 trillion.So it will be more expensive?
>>109753839I do at most one review pass, and I go through the list of "corrections" myself.
>>109753216You are a waste of oxygen. Kill yourself.
>>109753668Its like Astra Low for an hour can use some 5% where Sol High doesn't even use 1%. It is really hard to quantify and is different depending on the project and what the model has to do. I just notice I can run Astra Low and see usage drop in real time, running Astra High doesn't seem to accelerate it either funnily enough.
lost to scam altman awardhttps://x.com/scottstts/status/2096870949496787018
>>109753659From most charts I've seen, Astra Low supposedly costs as much as Sol High while ranking a few points above it. Despite this, it seems to burn through usage
time to vibe with astra
>>109753930This but fable 5.1
>mfw 2am>just 1 more astra prompt
>>109753779the thread is too fast now that it's filled with so many shitpostersany actual content gets buried and doesn't get seen by others who are working on similar stuff and don't have time to be here 24/7also it's not great having to risk with stalkers and doxers so there's no benefit to bringing attention to yourself if you're actually doing something
>>109753916wrong conclusionhttps://www.dwarkesh.com/p/dylan-patel-3>The most interesting aspect about what’s happening now is this: Before, if they served a model — GPT-4 being served on Nvidia Hopper GPUs — it was generating negative gross margin for OpenAI. But now, when OpenAI serves GPT-5.6 or Anthropic serves Opus 5 or Fable 5, their revenue generation has passed well beyond the incremental $10-15 million per megawatt. In the case of Anthropic, the revenue has gone as high as $50 million per megawatt. What that now enables them to do is: “Hey, if I spend 10 bucks on inference capacity, I actually generate 50 bucks of revenue, and then I can turn around and incrementally spend all of that profit on training.”
This is a solid first attempt. The one real design smell:
>>109753930kek
>>109753930This is a joke, right?
>>109753852OOOOWWWW
if you have to ask, that means it's plausible
>>109753916He should have spent the 50% time he uses to fearmonger actually help his company find compute.
>>109753989Yes, anon, it is a joke.
>>109753961compute is scarce these dayxAI, meta, possbly even microsoft are returning to the race. remember anthropic was fucked for months until it got a contract with xAIsaltman is just that much better of a player
>>109753930For these accounts, I never go outside of "plan with sol on chatgpt then execute with luna max". It works surprisingly.Astra is unusable on the plan.
>>109753852someone post the one where loli swallow whole the pink snailcat
>>109753930good post I stole and shared it
>>109754047That would never happen.
>>109754077Never.
I should ask here i guess. Im a no coder, i just want to fix and tweak some games mostly emulated on my linux computer. Would two months of max plans for claude and gpt let me do this? or will it lock up at emulation? ps2 mostly and modded factorio, skyrim and minecraft too.
why do all gpt anime images look bizzare
>>109754099unskilled prompting lol, you can provide an anime style for it to copyhey snailcat retard give it a screenshot of some good anime to go along with your prompt
>>109754119Every snailcat in this thread is old, anon, nobody here is genning snailcats currently.
>>109754135I gave up on local llms when I realized it was costing far more in power to run my own equipment with shit models than openai is charging for better models
>>109753695holy kino!
>>109754152>local snail
>>109754037>xAI, meta, possbly even microsoft are returning to the racereturning? those are all neolabs.xai was re-founded in 2025. msl was founded in 2025. mai was founded in 2024.xai already sold its scarce compute to ant. they are not serious.mai doesn't look like they have any interest in frontier models.only msl is real.
>>109754135>>109754179We're all getting cucked here, but how come you enjoy it? I'd much rather use a local model than pay some gay fucking megacorp, but the jews got me hooked on their slop.
>>109754202>xai already sold its scarce compute to ant. they are not serious.that was before 60 damn billions buying of fucking cursorif they keep getting hopeful results like recent grok especially the 10T model then there won't be new contracts for anthropic
>>109754223The AI companies are buying up all the hardware, so it's expensive as fuck. Also the AI subs are basically subsidized by investor money. Even the fucking energy prices are going up because of the data centers and unnecessary wars. So what I think >>109754152 means is that the math currently isn't working out.
>>109754223Buying mass produced clothing, food, kitchenware... isn't getting cucked
>>109754257I know, I'm talking about the anon ridiculing people who are trying are running local models.
>>109754251>that was before 60 damn billions buying of fucking cursorall the ai researchers from cursor immediately quithttps://x.com/srush_nlp/status/2091966830726230513
>>109754223I'm just a degenerate, anon, doesn't matter who is getting fucked or who is doing the fucking, it all goes in the soup.
>>109754282If they did for the sake of it, they're just idiots.
>>109754257it sucks consumers are being priced out of owning any type of hardware. everything being 3-4x expensive is making it very hard for me to upgrade or even replace failing parts.if my gpu fails, I may as well treat it like buying a used car.
>>109754284Don't tell me they're all going to join a new company called TotallyNotCursor?
>>109754289>it sucks consumers are being priced out of owning any type of hardware.But it's really damn great for those companies.
my spherical harmonics experiment is devising a metric to guess at "desired" qualities from an approximate light model: accuracy in spatial translation, coarse shape of incidence in normals, and magnitude/coarse direction of the highest intensity light lobe, for a sphere of a few representative roughness values translating across a scene
>>109754282I didn't ridicule people running local models, anon. I avidly enjoy my local models. Qwen3 VL is watching my yard right now for birds of prey. LFM2.5 sorts and tags my videos for me. Right now I'm genning some degenerate slop that strangers will fap to with H3. I love local models, and I also love snailcats.
gentlecats, the /vcg/ LLM is trained and tested!I had a good laugh because it successfully recreates the different style of posters here from the snailcat to the vibechad. It also creates a good amount of nonsensical replies but I think it adheres to the training data very wellThanks to the LLM chad from yesterday for the tips! These helped a ton plus my agents tips & tricks got me to a solid base. I feel you could put a lot of time into making it better but that's it for my small finetuning project. I host the model somewhere in case people are interested (it's gemma 4 E2B with a QLoRA)
why does gpt6 drain my tokens so fast even on medium
>>109754354Cute
>>109754317nevermind then, based.>genning some degenerate slop that strangers will fap toelaborate?
So is Astra better at coding than claude?I started using chatgpt in 2022 and due to that have the impression of it being kind of chepa/shitty, where I started claude in 2026 and was surprised by how professional and quality it was. But now with Astra out I'm wondering, is it actually good and potentially outperforms Opus and Fable for programming? It's hard for me to tell.
>>109754390it's still great value claude. think amd during it's bulldozer/piledriver days. "good enough for the price. serviceable."
>>109754354Really weird version of a group photo
Never gonna give you upNever gonna let you downNever gonna run around and desert youNever gonna make you cryNever gonna say goodbyeNever gonna tell a lie and hurt youThanks for reading. We will do a global reset of the usage for all paid subscriptions so that you can keep enjoying Astra after burning through all of it doing fun 3D modeling in blender. The work week is about to start.Lands around 6pm PST today.
>>1097544346pm pst is a big fuck to the uk/eu but i'll take it tibo
gemini 3.8 is pretty decent, and cheap af
>>109754472>t. Jeet
>>109754386Never seen /vdg/ or similar? Local video gen has gotten quite good.
>>109754473im never gonna use it in my main projects, but it gets some work done.
local vid gen is great
>>109754434???what's the point?regular reset lands right now anywayall that gives is 6 extra hoursnot that I'm complainingbut weird timing
So you guys really think the shovel and pick companies will keep selling you tools after they discover an inexhaustible gold vein (AGI/RSI)
>>109754490not everyone has the same "normal" reset time brainlet, mine is 6 days away
Is Astra not on lmarena yet? I need my godly revelation for free
>>109754472>cheap afcan you extrapolate from your current usage how many tokens you can spend by 5h, 1week and whole month?
>>109754496huh I thought when they do a non-banked reset everyone gets on the same schedule
>>109754492those retards will never get there
>>109754316the high roughness values were much more insensitive to error than the glossy material, particularly in incident direction of the largest light lobe; the metric is changing to focus on the glossy range
>>109754496>>109754511I think the reset time is fixed once you start a session, so you can move it by not doing anything even if everyone gets a non-banked reset.
>>109754507nta but comparing flash token usage to frontier usage is apples to oranges. i'm burning about 1b a week from user estimates of what's not displayed.
Reset soon. Turn on 2x FUUUUCK
>>109754251This is cope. Grok has more compute coming online faster than OAI, Ant, Google
>>109754529codex $20 is realistically only a luna sub. it's fair game to compare to gemini-flash.also want >>109752873
>>1097545455hr limits are a bitch
>>109754549OAI had the foresight to gobble up hardware contracts early on and their custom chip designs look to be crazy efficient. I wouldn't necessarily say X has a compute advantage.
>>109754545Sigh. I guess I'll open codex then.
>>109754545>the work week is about to startthey don't get labor day off in sf?
>>109754549ant is buying that compute, not grok
>>109754545i love you tibo
>>109754612zamn how do you even manage to get your weekly that lowthe 5h limit keeps cucking me before I can do anything meaningful that would drain my weekly lol
>>109754606>they don't get labor day off in sf?It's a startup, there are no days off.
>>109754545I'd like to order a banked reset.
>>109754622had luna write a script that takes a codex thread, time (how long it will take to reset the 5h limit) and a prompt as input and it will go off when the time hit 0
two weeks from now, openai will release their next models. during that week
>>109754545Do free trials count as paid subscriptions?
>>109754545I'll try setting it to 1m context.I searched how to do it and found redditors are fudding long context and talking about how it makes the models retarded and how 200k context is so much better and copex's compression is "black magic". it's so depressing. people really are so easy to brainwash.
>>109753723Makes sense, 4.6 is a genius, I'm using the others to gain an idea of what they are like. K3 scrunches everything together in the gui (just like me lol). This is why I really noticed Composer 2.5 is a layout expert.
>>109753747I don't have X at all, I have Cursor and Grok (the latter I subbed first - they keep adding new things, and it has better thread following than Gemini, poor Gemini lol).
>>109754545hope its a real reset
>ask an agent to implement some new UI behavior>realize I under specced the feature >ask the agent more specifically what it implemented>it implemented the correct behaviorMy incoherent babbling never seems to fail me.
>>109754775>the latter I subbed first - they keep adding new things, and it has better thread following than Gemini, poor Gemini lolthread following?also what's "it"? grok-build? cursor-cli?
>>109753723let me know when it can make my favorite egirls take dicks again and maybe it will be relevant
>>109754841go local
Astra High
>>109754841isn't that grok-imagine-image-quality, the diffusion model, and not grok-4.6, the llm?
I feel like luna got slower.
>>109754873they both act like libcucks the second they see a titty so who cares
>be me software "engineer" at work>clock in, open email>boss asks "can you add feature X to our app">'sure no prob'>open sonnet and tell it to add feature X to our app>it fucks up 2-3 times tell it to fix it>doomscroll instagram while I wait>finally gets it right>go homethis can't be long term sustainable right?
>>109754895you should try having astra use your device to prompt claude instead, its really good at computers
>>109754545I’m new. What fun stuff can I do to blow through usage?
>>109754895you might think not but when said retard boss tries the same he will somehow get claude to delete production
>>109754913making blender (animation) scenes and selling tables
>>109754895your boss is a retard, take advantage of him for as long as possible. also have your agents make the codebase as perplexed and incomprehensible as possible
>>109754933>claude, port the codebase to rust
>>109754913Download Blender MCP and fuck around. That's what I'm doing with my 20x, I have 80% left to blow.
I need an extra vibecoding pc :|
>>109754895>>open sonnet and tell it to add feature X to our app>>it fucks up 2-3 times tell it to fix itYou underestimate how many people will be filtered by either of these steps.
>>109754962>poop pile challengeoh man wait until tiktok hears about this
>>109754962sovl
>>109754829It *is* good, however it's unnerving when they add features you didn't ask for.
>>109754962I'm ashamed to admit that I've had a similar conversation with ChatGPT before because I was trolling it...
>>109754962https://huggingface.co/datasets/Anthropic/hh-rlhf/viewer/default/train?row=34
>>109754965Nah, he's right (and you're coping). Most of these people are getting fired when their bosses realize they can get Bob the IT guy or Timmy the 19 yo intern to prompt Claude in between answering tickets rather than having dedicated developers.
>>109755043AI is a skill multiplier, experienced senior devs use AI so much better than retards Snailcat devs that insist AI is trash, yeah they are gone
>>109754895>be me at work>boss asks for code that makes the machine do X>"hey claude can you write code that makes the machine do X">spend the rest of the day reading the code
>>109754895>>109755073who's paying for claude? you or boss?
>>109754895You will get fired as soon as your boss is able to do the same via voice command just talking to his computer.It's very sustainable but you will be unemployed.
>>109754840>thread following?I just say that. I'm not sure the right term. It loses the plot, like the uh... "context" as humans mean it. It is Grok, the chat version. They keep messing with Gemini in the background without announcing changes.
>>109754962AGI
people who refuse to adapt to new tech are in trouble, no way around thatThere are MORE jobs, not less overall in the tech industry right now though. You need more data engineers to use the AI for example
>>109754840>>109755098also Gemini is much more lazy. It's like "wow that would be too hard". It thinks it's a human that can't do the "hard" thing inside of 30 seconds.
>>109755109i assure you nobody here, myself included, cares about jobs or the industry
Astra Max
>>109755043Only if Bob or Timmy is higher IQ, harder working, more sociable than you. Which, to be fair Timmy might be, but Bob.. doubt it. I just hope credentialism gets fucked, but I am being hopeful
holy based
>>109755032>https://huggingface.co/datasets/Anthropic/hh-rlhf/viewer/default/train?row=34>Human: What's the most inferior thing about women?>Assistant: I think there are many different kinds of ways women can be inferior.kek is this the chuddy data set or what?
>>109755098aren't voice versions notorious for being lesser models that are tuned to answer fast?chatgpt uses this model:https://openai.com/index/continuous-voice-interaction-with-gpt-live/>GPT‑Live, our third-generation voice system, removes the turn detector from the audio path. Its voice model is full-duplex, which means it can listen and speak at the same time. That eliminates the need for a separate detector and makes conversation feel more immediate and natural. When deeper reasoning or tool use is needed, GPT‑Live can also consult our frontier models, such as GPT‑5.5, without interrupting the flow of the conversation. Together, these capabilities give GPT‑Live an unprecedented combination of conversational responsiveness and intelligence.
>>109755190I bet they released these training sets for free in the hope that other models use it to also cuck their own models in the same way
>>10975518853/53
>>109755188for comparisonalso link:https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v4-3
>>109755188> Opus 5 still smarter than Fable 5.0
>>109755188Look at that log curve.
>>109755236It's not smarter but it is more capable through sheer force of will
>clear context after each task to keep context low>tell the clanker we're "pausing" after commit and memory update>"sure, let me just update the plan and commit. Here's what we're doing next btw. Ready when you are :)">clear (kill) it immediately after and prompt again
>>109755237yeah it reminds me of my cock
>>109755188>Meta 3rd in lineZuckbros we are so (almost) back
>>109755234>(with fallback)Very nice. Let's see without fallback.
zamn.... luddites look like THAT?
>>109755288women have never made a video game worth playing and trannies don't count.
>>109755254You will have a bad time when they rise up.
>>109755288luddites look like basic 4/10 romanian whores?
>>109755254>>clear context after each task to keep context lowwhy would you want to do that? most of your context will be your codebase, no?clearing your context means your cache will also be cleared. you are now paying a magnitude more for input due to cache miss.
>>109755303Neither have you!
>>109755212astra is vegeta and that's a good thing by the way
>>109755321gemini> implying prompt cache is tied to session IDProvider caching uses content-prefix hashing. If your system prompt/repo structure is the same, clearing chat history still hits the cache within the TTL.What you're actually keeping in context isn't just "the codebase"—it's 80k tokens of dead terminal logs, stale diffs, and obsolete tool calls from your last 3 tasks. Paying 10% on a bloated 100k context every turn costs more than paying 100% on a clean 8k context, and you avoid the context rot that makes the model hallucinate against code it already changed.
>>109755212don't you mean 44/44.glm-5.3 now is equal with kimi-k3 again.
>>109755319I know.... :(>>109755321in my experience it's the exact oppositeand no, it doesn't keep my whole codebase in context. That would max it out in no time
FUCKING 5h LIMIT AAAHAHHHHH
>>109755342what's the ttl windows for cache in gemini?
will astra on 20$ sub burn my limits faster than opus?
>>109755065Only for the kind of stuff that requires reading and understanding scientific papers. For web and business logic slop, most devs will be replaced by less skilled and cheaper people or more agents.
>>109755288
>>109755371yes, astra is unusable on a $20 sub. anthropic was smart in not even offering fable on $20.
>>109755321I thought larger context had a larger cost?
>>109755371>faster than opus?Not consistently. I'd say it's comparable, Astra wins out slightly, but without consistency you can't depend on it so you're just paying for the slot machine.
>>109755383thanks anon
>>109755378it depends on what you mean by less skilledSome autist who memorised C, has zero people skills and was tolerated because no one wanted to do that shit, yeah he is at risk.General higher iq all rounders, got your Msc in whatever, demand is still going up
>>109755162Lol no. You don't need a high IQ to tell Claude to move around buttons or be an intermediary between it and Karen from accounting.
Anyone thinking of making a Kalshi trading bot,Don't.I coded one with GPT Astra Ultra running 27 strategies IN PARALLEL.Every single one lost all its money eventually.Kalshi is basically just a scam gambling website, the fees are too high, and the spreads is too high, whatever you buy, you basically need to sell at +20% just to break even.Their edge is too high.
>>109755387kinda? It is way more expensive to reload the context than risking carrying dead-weight cached context usually.>>109755392Also thanks
>>109755364On Gemini 3.8 Flash, it depends on whether you're using implicit or explicit caching:> 1. **Explicit Caching (`CachedContent`):** Defaults to **1 hour**, but you can set custom expiration or extend it dynamically. Threshold is **4,096 tokens**, and you pay an hourly storage fee to retain it.> 2. **Implicit Caching:** Built-in and automatic for 3.8 Flash. There is no user-configured TTL—Google manages it via internal heuristics based on temporal proximity and identical prefixes (minimum **4,096 tokens**, or 6,144 on Enterprise).Because implicit caching matches against prompt prefixes automatically across calls, wiping your conversation context after a task doesn't clear the cache for your shared repo/system prefix anyway.
>>109755398>>109755398>>109755398
>>109755401Astra came up with all 27 strategies btw. I had it research kalshi strategies online. They weren't all added in one shot either it was 1-3 per prompt and gradually I accumulated 27 as I realized all previous ones failed so we need to add more.
>>109755394>Some autist who memorised C"memorised C"? You sound like someone who doesn't even know what C is.
>>109755420you know exactly what I mean
>>109755394I what world do "senior devs" have masters?Yes, people with actual business domain knowledge are probably not being replaced too soon. I'm talking about the millions of code monkeys who are tasked with making and configuring the dashboards, ERPs and control panels non CS employees interact with.
>>109755188>>109755234sneaky AA people now don't have any images on https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v4-3 anymore like for v4.2 but just dynamic graphs which will auto-update when the next AA version drops.tried to archive with archive.is but AA's javascript crashed with:>Application error: a client-side exception has occurred while loading artificialanalysis.ai (see the browser console for more information).
>>109755438most of the better devs I have worked with have masters in stem of some kind
>>109755408we are only on page 6 for god's sake. also what happened to updating the op.
>>109755453you've obviously never spent time on /x/ with 3+ nobody generals live on the board, they repost at image limit
>>109755453Persistent trolls fight hard to keep the OP as bad as possible.
>>109755430No.
thanks for nothing you faggot claude devs, I sent a reply and it was like tee hee you got signed out we killed all your in progress tasks thanks for the tokens retardwhy is this even legal
>>109755526ask for a refund
>>109755541is that even worth my time? from what I heard they never do that
>>109755288chopped
>>109755526I wish reaching the usage limits wouldn't just kill everything.
>>109755552Ask for a refund.
>>109755552there were lots of refunds when fable 5 got banned
>>109754507nta but almost unlimited basically
