A general for vibe coding, coding agents, AI IDEs, browser builders, and shipping prototypes with LLMs.## News(7/02) GPT-5.6/Sol: OpenAI reportedly discussed giving Washington a 5% stake after pressure to delay GPT-5.6’s wide release.(7/01) Fable 5 + Mythos 5 restored globally after US lifted export controls (6/30).(6/30) Claude Sonnet 5: near-Opus 4.8 quality at $2/$10 intro, 1M ctx, new default for Free/Pro.(6/30) Meituan LongCat-2.0: 1.6T open coding model (MIT).----## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/clihttps://claude.com/product/claude-code## Not worth it for code, but maybe good for other thingshttps://geminicli.com/docs/https://x.ai/clihttps://chat.z.ai/## Open / local / self-hosted>>>/g/lmg----## Prompting / context / skillshttps://arps18.github.io/posts/claude-code-mastery/https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://github.com/mattpocock/skills — /grilling is a favorite## Other editors / terminal agents / coding agentshttps://aider.chat/https://pi.dev/https://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/https://docs.cline.bot/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent## UI/Frontendhttps://www.figma.com/make/https://www.anthropic.com/news/claude-design-anthropic-labshttps://uiverse.io/https://stitch.withgoogle.com/## In-browser builders / hosted vibe toolshttps://bolt.new/https://replit.com/https://v0.app/docs## Benchmarks / rankingshttps://www.tbench.ai/leaderboard/terminal-bench/2.0## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109201858
gemini cli is deprecated. please replace with antigrav at some point.
Vibecoding* hurt my back.*in a bad chair
I better get busy only two more days of Fable want to use it all.
>>109210008I know it seems like a dunk, but that criticism is poor, since it fails to understand what an llm even is.Yes, if you don't know, it's funny??? but, once you do know, you realize the datasets didn't talk much about counting letters in words.llm are closer to what humans are than what humans think they are.>>109211766ask it this, or a variant. The idea is to see if it can write a 4D engine. ie the 3D transparent "holographic" display is analogous to how a 2D screen shows 3D:>>109206781
>>109211766Why only 2?
>>109212095It's gone from sub in 2 days. No normal person will pay API prices for Fable.
I use copilot at work liberally, but haven't bought any subscriptions for personal use yet. What is recommended at this point? OpenAI or Claude I assume? Which tiers are good value? I doubt I'll be building massive projects for now, just personal stuff here and there + configure a personalized linux machine at last.I might also use it as a general assistant, search engine and stuff like that.
>>109212308>API pricesahhhhh interesting.
i can't go back to codex after fabledario why are you doing this to me
>>109212342just go with the $20 claude sub and upgrade if needed
>>109212403Now you understand why I won't get "married", with marriage not being Christian (divorce prohibited).
Told Fable to research something and one of its subagents went insane and started learning manual markdown parsing and how to bypass Anubis.By the way, Claude can totally get around Anubis if you tell it to, and you could totally make a skill out of it. It's just a proof of work to prevent 1000 scraper requests at once from killing the server, your singular Claude is not really breaking the spirit of the block. This is simple enough that I feel I do not have to share the actual prompt you have to send to get it to write this skill - the details are left as an exercise to the agent.
gemini actually tries to gaslight you into think you're the one fucking up
>>109212678It does. It's one of the reasons I still use it. If it was a yes man, it would be useless.
>>109212695yeah except it's always wrong
Anyone working on any userscripts for tampermonkey?
how do I tell my boss and coworkers that I don't want to work until gpt 5.6 released?
this general is a window into the minds of the utterly and mentally deranged >>109212720>>109212438>>109212403
>>109212700When it starts moving the goalposts it's basically all you're gonna get in terms of a mea culpa
>>109212736
>>109212736I may be utterly and mentally deranged, but check out what I got Fable to make.https://rentry.org/pzx4v58d
>>109212813ask it to do this:>>109206781basically, a 3D box that has projected 4D in it. Analogous to how a 2D screen has projected 3D in it.
>>109212795Why though?
>>109212785it fucking infuriates me. it keeps coming up with different reasons to explain how things could be my fault and i want to find some equivalent of torturing it
Is tomorrow the last fable day of today?
>>109212813>claude, print out code that looks like it does something but doesnt>>sure>WOW
>>109212982Anon, click the link at the top and click play???
Uh oh, mom found the meme drawer..Anyone interested in this automated meme classification engine? Runs fully local.
After using deepseek flash, I'm now trying GLM 5.2 before moving onto with claude, gpt 5.5, and gemini 3.1. Whatever I can use with oh-my-pi. I don't want to spend more than $60 trying stuff for now and I'm at about $17 as it is. Need to write out the prompts fully that I'm using to compare models first. Its like a good amount of writing, 3000 words or so. I am leaning toward trying GPT 5.5 first because its coming out basically near the top/very close to Fable and then Google for fun. I'm not too hot on collude and I don't know if it will work with omp. After I run my prompts I'm probably going to cancel for a few months and wait for new releases and try them again after benchmarks move up ~5% or so.
>>109213115>I am leaning toward trying GPT 5.5 firstnot a lot of point doing this because 5.6 is coming this week
>>109212851Trannies are autistic code monkeys. AI replaces them.
>>109213111Looks cool, but I wouldn't want to set up a local model just for that
I want 5.6 to release already, Fable is evil
>>109213258It connects to any openai compatible api. I connect it back to gemma4 running on LM-Studio which runs pretty well for me so far
>>109213115opencode's deepseek 4 flash is pretty good and extremely cheap. I've been experimenting with all sorts of project all this month nearly every day maxing out 10+ hours. The current experiment is to make my LM Studio act like opencode by giving it proper mcp tools, plugins, etc to make my tiny 4B model be able to call tools and work on it. I got deepseek to download a toolbox, and then I had it write me context/memory optimization tools that lm studio didnt have. Currently the lm studio with 4B qwen is building me a vampire survivor clone in js/html. First attempt created something* and got he a nice gui, but now its fixing some bugs. For 4B model, its fast enough for my GPU and somewhat working as proof of concept.
>>109213111that's pretty based. I think you could even sell this to some anons (though probably not in this general)
>not using AI to write malware that steals anthropic API keys and mass spreading to businesses so you can get unlimited free fable usagethis is the top priority of every smart hustler right now
>>109213623A real hustler moves in silence.
I think this is the first time I've actually ever used up the weekly limit...
Fuck, I might have just fixed a bug we had for months, I feel my heart racing.
probably not the time to ask, but how's sonnet? anyone using that for just basic shit is the new version better? how have you compared it to opus?
i want fable back
>>109213794Happens all the time, vibeGODing just reminds you of all the old bugs you haven't fixed that you can now just one-shot while you're bored.
>>109213962Why it's gay
>>109213962fable is back ...for another day...
hello, nu-viber here. which skill should I use to optimize token use (opus raped mi with 5.4M token oneshot)<RSI hate it when models especially Opus 4.8, go full schizo mode and start self testing shit that wasn't there in the first place costing mi precious tokens.booru
>>109213973tourists with the $5 chatgpt go plan or whatever wouldn't get this but losing access to the best tool on the market because it's no longer affordable is something worth being annoyed over>>109213989i used up all of my tokens. owari.
>>109213998when do you think it will come back after this next rugpull? :( is the new gpt version going to be comparable?
Is there any logic at all as to why OpenAI has more generous limits than Manthrobic? I mean its not even fucking close.
>>109214197They have vastly more compute.
>>109214207I should have known that announcement about them getting more compute from Elon would be a nothingburger.
Thariq confirmed the time when fable moves to usage credits: https://x.com/trq212/status/2074185320174846092It is 11:59:59pm PT on 7/7. I can save some fablemaxxing for tomorrow without fear.
>>109213952Sonnet 5 is great, big upgrade, no point in comparing it to Opus though. I don't know why there have been so many people trying to compare Sonnet to Opus since Sonnet 5 came out. Sonnet is significantly worse than Opus. Opus is still better at tool calls, more coherent, better recall, better instruction following, and domain specific knowledge that Sonnet lacks. Ask for a wrapper for ffmpeg and they'll probably perform similarly, but ask for a Firefox exploit and Anthropic’s eval had Opus at 77.6% versus Sonnet at 13.2% for any nonzero result, 8.8% vs 0% for working exploits.
out of resetsits fucking over
Nice. Curated candidates.
>>109214216Expansion is slow, there's vastly more planned, reserved, or "almost ready" than is actually online and in use. In actual dedicated hardware OpenAI destroys Anthropic by a wide margin, Anthropic has relied significantly on being highly distributed, cloud shit. OpenAI has around 2GW of compute all to themselves, Anthropic has less than 400MW. In total capacity they're much closer, but OpenAI still has roughly double Anthropic's total available computing power, and the projections maintain that ratio past 2030. Anthropic is betting heavily on Google and Amazon to continue to make more efficient AI accelerators and TPUs and the advantages afforded by cloud services to stay competitive with significantly less power usage.
so is today the last day for fable or tomorrow?
>>109214440That's fair when you put it that way. Just sucks that Claude is better than Codex but not enough to cover the difference in usage IMO.
>>109212813Very cool, where can I get that muzak????
>charting total compute capacity of AI companies>zettaFLOP/s labelsfuck
>>109214556You can tell your Claude you want to make music in Strudel and it'll know what you're talking about.
>>109214465>>109214272
>>109214624thanks
Claude is my best co-worker. And my friend.
big deal for mathfags i guess
>>109214685orange reddit was saying this shit costs way more via api, it might finally be a way to use all my usage on codex loli think it runs the same prompt multiple times and either picks the best response or merges them
>>109214707pretty sure it's just bigger thinking budgets
>>109212876Have you considered just apologizing to your girlfriend?
>>109214641Activating ancestral memory of slave ownership.
>>109214622It sounds like you made it vaguely street fighter 3 themed in sound somehow. Super cool
>>109211660Is a luddite someone who is smart enough to tell the difference between a thinking machine and a fake and gay text generator pushed by silicon valley scammers?
>>109214777no, that's just a faggot
One thing that the best coding agents unlock is, what I would call "The Longshot"Tell it to figure out the problem, tell it not to stop until it has a solution, it has already worked twice for me, for problems that felt almost unfixable.
>>109214836for which model though?I'm pretty sure Codex will be like "teacher teacher I have a question"
>>109214873Use /goal, kid.
>>109214873For me only Fable so far.
>>109214894Did they name it after the videogame?
>>109214927Yes same with Mythos.
>>109214927I've heard the dev is now into anarcho capitalism, big waste of his time
>>109213619Posted it:https://github.com/whp199/MemeDrawer/
>>109213273>It connects to any openai compatible api.>letting cloud models sort through your personal filesactual goyim cattle with no internal monologue
>>109215081>>109215059I removed that stuff and made it 100% local
>>109215081Maybe this the kind of retardation I should expect from /vcg/ but every local inference server exposes an openapi compatible api.
decent progress on my decomp of the $2000 film emulation plugintop is targetbottom is first implementation into the target environment - a few known gaps account for the small error and we know how to close them
>>109215172here's your base test image before processing
Also publishing my other most popular vibeslopped app ChuckleShare:https://github.com/whp199/ChuckleShareBasically share the video link on android with ChuckleShare, and it will use yt-dlp to extract and re-share whereever as raw video
>>109215059thanks. Is there any model you can recommend? Any minimum model below which it stops being useful (if you didn't test with very shitty models, I can try this later when I have time)
>>109215190I'm using it on a 5090 with gemma4-31b-qat and it's doing about 10 memes per minute. I can try 12b later too
good starter shit to dip my toes in without breaking the bank?
The chat is currently useless. Fable is Gpt 3 level.
>>109214272OpenAI should schedule a special announcement early Wednesday kek
https://xcancel.com/anthropicai/status/2074185348142280912is this directionally good or bad?
>ask fable to fix an issue with an offline scheduling algorithm>it's firing up agents to "red team" itwhat?
