A general for vibe coding, coding agents, AI IDEs, browser builders, and shipping prototypes with LLMs.## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/clihttps://claude.com/product/claude-code## Not worth it for code, but maybe good for other thingshttps://geminicli.com/docs/https://x.ai/cli## Open / local / self-hostedhttps://github.com/OpenHands/OpenHandshttps://github.com/QwenLM/qwen-codehttps://github.com/QwenLM/Qwen3-Coderhttps://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF----## Prompting / context / skillshttps://arps18.github.io/posts/claude-code-mastery/https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://github.com/mattpocock/skills — /grilling is a favorite## Other editors / terminal agents / coding agentshttps://pi.dev/https://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/https://docs.cline.bot/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent## UI/Frontendhttps://www.figma.com/make/https://www.anthropic.com/news/claude-design-anthropic-labshttps://uiverse.io/https://ui-ux-pro-max-skill.nextlevelbuilder.io/https://stitch.withgoogle.com/## In-browser builders / hosted vibe toolshttps://bolt.new/https://replit.com/https://docs.github.com/en/copilot/tutorials/sparkhttps://v0.app/docs## Benchmarks / rankingshttps://www.tbench.ai/leaderboard/terminal-bench/2.0## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109102722
/compact
Nobody vibecodes anymore I see.
>>109112732AI took my vibecoding job and now I'm in the permanent underclass
>>109112732We're all waiting for Fable to come back.
>>109109221>>109110433Did some more improvements on the pipeline side, haven't looked at any more kernel/model side optimizations. Running it on MKULTRA docs now, ~17k docs/pages, I'm seeing ~6k per hour with longest side 1280 and max new tokens 1024Pretty good desu
>>109112786>We're all waiting for Fable to come back.Nope the jew took it down. It's not happening.
Yawn. Where are the new models?
>>109113220I need Qwen4-20B
>>109112584FPBP
guy at work told me he used 5 million tokens with claude as if that was a lot
>>109113560Wow five whole contexts?
>>109113560kek I used 500M last week alone
>>109113560lol. I'm not even a power user and I've used 3.5B in the past month.
>>109113560
>>109113617> Longest task> 53m 50sheh, I almost got to the 3 hour mark once. Get to my level
>>109113560I am wondering, I am quite late to the game.I have used OG ChatGPT in 2023 and then was out of the game for quite a while until mid-late 2025 when we already had 100-200k context windows readily available.How was vibecoding in 2024 and early 2025? Didnt models used to have like 30k context?
>>109113617so what you're saying is you hit the daily token limit within an hour LMAOimagine what you'd be doing with local AI and unlimited tokens, that number could have been in the trillions by now
>>109113667why have limited bullets when u could have unlimited crayons lmao
>>109113667>trillionsI'm writing code for an MCU so It's not easy to automate the testing because it's for a HID. I have to validate every change manually so the tokens last me through the week.
>>109113698fable would have written an emulator for the MCU and used it for static testing
> everything works in localhost> deploy> nothing works anymoreFUUUUUUUUCK
>>109113720GLM5.2 would find, download, and use an existing emulator for the MCU and use that for static testing.
>>109113819have yet to see anyone make anything or even talk about making anything with glm
>>109113832I've been enjoying it, but I mock the shills who try to compare it to Fable. It's easily 5.4 / 4.6 tier which is impressive for ching chong AI and plenty useful, but that's it. I've been using it for Medium and Small tier agents in distributed workflows to save on GPT5.5 usage and I've been happy with it.
>>109113737give it access to prod, anon
Guys they removed Fable from my sub, what should I use now?
>>109113890Opus and ChatGPT
>>109113890GLM 2.2 with OpenCode
vros I'm talking to vanilla gemini about AI companies and subscriptions, and this is the SECOND time it's suggested itself lol. Thanks gemma, ur a peach.
>>109113890codexgroid... nobody has had access to fable for over a week..
>>109114031Worked on my machine.
>>109114006THIRD TIME
>>109114078At least it's actually offering DeepSeek as the number 1 choice and itself as the humble second option
>>109114078And a fourth time.Starting to wonder if giving Bezos my money might actually be the right idea...
okay, it is as I feared. AI is gay.
It's a pain having to feed a whole file to translate in small chunks to not get over the token response limitAre there no free automated options? I ain't spending money on tokens
>>109114175claude code is $20 get a job kid
>>109114097Get a Claude to review a script that calls an LLM and there's a fair chance that it will change whatever model you are using to Haiku or Sonnet.Get GPT5.X to review a script that calls an LLM and there's a fair chance that it will change whatever model you are using to GPT4.I have noticed it less with Gemini though.
>>109114197I'm not paying to make something I'm going to give out for free after I'm done. It's a matter of principle
>>109114208I have decided that I hate Anthropic and OpenAI as companies and won't do business with them until shit changes. I'm just sorta fishing for models right now.
>>109113692worked against imperial japan in ww2, also nice of you to admit you cant afford a single 5090.>>109113698oh I see, someone is paying you to do this, well in that case good job? You'd be even better at fooling your employer if you had unlimited tokens
>>109114222like, to do everything I want for the price I want would take a setup that is polling multiple different companies' keys for different tasks and quite frankly I'm just too tired at this point to deal with that.
>>109114222The only remotely comparable model is GLM5.2 and it's still well behind the options from OpenAI and Anthropic. Everything else is much, much worse.
>>109114238brother I'm using glm 5.2 and it punishes you for doing anything other than code with its subscription. Like, asking an agent to order a pizza with a skill. You get throttled or banned. API keys only, and their API is shit for basically anything but large multi-file projects. It's one model for this and one model for that and some can be subbed and others will rape you with an API key. It's just so fucking shitting kiking tiresome.
>>109114266You're describing issues with Z.ai as a provider, not issues with GLM5.2, and I've spent the last week telling people in every /vcg/ to NOT use Z.ai as a provider because it's so bad.
>>109114273Grok, explain to this man that I'm sick of dealing with provider issues
>>109112584Compactfags are kidding themselves. This is a /clear board.
>>109113656Models with 1M context windows are an utter step change in performance over the 200k context windows. Went from toy to tool.
>>109113737>format laptop to the exact OS as my prod server>deploy>it just werks
>>109113656You went to jail?
/plan/goalI sleep My clanker will have something good for me when I wake up
What VS Code extension do I use to support OpenRouter chat?
>use gemini 3.5 flash (medium) via anti gravity ide>got google ai pro from a phone promo>goes in circles over the same problem regarding mirroring of 3d models>burns up all tokens>try sonnet 4.6 with same implementation plan (for some reason included in the google ai subscription and on separate tokens)>solved in 5 minuteswtf is google doing
>>109114937>can't keep face designgrimAnyway, why's she on a playground?
>>1091149863.5-flash-medium is like setting a downy to medium. no reason to do that. for thinking 3.1-pro-high is still the best google model. and, yes, claude is better at a lot of stuff.
>>109112567>postan in this totally organic thread>totally not forced>totally not calcified in contradictionattack my character
>>109115036
>>109115043and it goes in the butt btw
>>109115002weird that it's explicitly recommended to set it as mediumhopefully they push 3.5 pro soon enough
Thoughts on the vibe-coded geography game I've been working on?
>>109115069>weird that it's explicitly recommended to set it as mediumbecause 3.5-flash eats about 3x-5x the tokens of 3.1-pro for the same task. weird but true.
>>109115036*hits this snailcat off the stoop with a broom*shoo!
you have been taking advantage of the codex token counting bug right /g/?>>109113560>>109113617been using 10-15B a day on average
>>109115116makes sense, it often seems to run in circles over complex tasks and burn a bunch of tokens figuring out things it knew just last prompt
>>109115036you on the left
>>109115185Another Bosnian classic
>>109115150>you have been taking advantage of the codex token counting bug right /g/?what's that, precious?
>>109112584> compact? more like cope.
>>109115109I like the map a lot and the colors of the map but I don't like the text it's difficult to read, can you vibe in some different themes like solarized etc that change the fonts and background color
>>109112732vibecoding was always cope for people who couldn't read docs
>>109115109Where's the Bing Bing yahoo!
>>109115312>I don't like the text it's difficult to readIt looks better in person. I'm using Libre Baskerville and IBM Plex Sans. Going for a 'mid-century cartographic-bureau publication' vibe.>>109115325>Where's the Bing Bing yahoo!I'm deliberately eschewing gamification. My AI agents are actually forbidden from gamifying through binding specs.
>>109112584I created a handoff command that I named /checkpoint.
How do you monetize vibe-coded AI stuff? The servers are costingI made a lot of slop which I want to sell, for a premium audienceSteam did not like my vibed game because it was too smart.I'm leaning on free desktop + paid smartphone app + begging(I'm actuallly a professional vibe coder, and I use Antigravity)
>>109115457I have a regular job that pays me money
>>109115457>Steam did not like my vibed game because it was too smart.What do you mean?
>>109115457
the general vibe ITT right now
It is my pleasure to announce that we will more than likely get Fable very soon, as soon as my weekly token quota is set to expire.>>109115457If you have a job, you can (responsibly) vibe code stuff for work. It can help you get paid while developing skills, or rot your brain, either/or.
>>109114212>I'm not paying to make something I'm going to give out for free after I'm doneYou'll pay with your time then. If you fork over $20 you'll get it done 10x quicker and with a lot less annoyance.
do I have to race the clock to beat cache dump and avoid being charged extra?
instead of ctrl+f'n, searching "keyword", find and replacingdoes the model, inject the entire context, predict where it should be, replace it?
