A general for vibe coding, coding agents, AI IDEs, browser builders, MCP, and shipping prototypes with LLMs.## NewsGLM-5.2 released, Opus performance with MIT license and low pricehttps://x.com/Zai_org/status/2066938937344495629Unreal Engine 5.8 ships today with experimental MCP server support:https://x.com/UnrealEngine/status/2067251500900839735----## 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## Mid tierhttps://opencode.ai/https://cursor.com/docshttps://antigravity.google/## 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 — /grill-me is a favorite## Other editors / terminal agents / coding agentshttps://x.ai/clihttps://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## Previous thread>>109074367
how's the vibes
>>109082331No vibes, only pain.
You think you the Fit, bitch? You not even the Quant (Grrah)
The free claude model is dogshit, is opus as good as people hype it or nah?
>>109082320>grill-meit’s “grilling” now: https://github.com/mattpocock/skills/blob/main/skills/productivity/grilling/SKILL.md
>>1090823424.6 opussy is actually a good model that's worth it. but the others... hmmm...
>>109082320Fake image. Snailcat would win the race after the others get cocky and stop to flaunt in front of the finish line.
>>109082368we were all snailcat five years agowe’re actually shipping things we couldn’t have shipped five years agoyou sound like you’re coping for not shipping anything ever
>>109082320sexo. sexo with snailcat.
>>109082368>Fake image. Snailcat would win the race after the others get cocky and stop to flaunt in front of the finish line.
I'm officially impressed with GLM5.2. I didn't expect it because I've spent the last two weeks hating on all the new chink models because they're benchmaxx'd, too expensive, too slow, too retarded, some are better than others but none are worth using. GLM5.2 is plenty decent and I actually think well worth considering if the price is right. It's been obedient, it's good with tool calls, lacks vision capabilities but it's better at UI than GPT5.5 and has a huge context. Genuinely not bad at all. Easily 4.6/5.4 tier. Good job, Xi.
>>109082404Better than DS?
>>109082410Vastly, I hate DeepSeek and consider it unusable. I still think if you're not spending at least $20/mo with both OpenAI and Anthropic there's no reason to consider other models at all.
_@^..^
>>109082430I've been poverty vibecoding with DS API due to the cheap tokens and it's a one off reverse engineering project so I didn't want a subscription. I still don't want to bite the bullet and pay a month of Claude so I'll check out GLM. It's a matter of principle, I don't want them to be able to turn it off completely anytime like Fable. Thanks.
>>109082439I have no idea how the pricing compares in real-world usage, I know DeepSeek is famously cheap, and I know the GLM coding plans from Z.ai are a very fucking bad value. It's still fresh enough that new providers are available every day, be sure to shop around.
>>109082368Snailfags are so delusional hahaha
>>109082472
FUCK bosnia lolAs useless as AI
...@(=^･ω･^=)
i made a few prototypes and games a few months ago and i want to jump back in. should i stick with claude? i was using a mix of claude and cursor at the time. that's about as much as i want to spend-- like 2 basic plans
>>109082477kino checked
>>109082531codexcodexcodex
>>109082531Claude and CodexClaude for UI, Codex for autismFable for telling what Codex to do if it comes back
>>109082546>>109082567perfect, thx
I'm in jail for 10 years and snuck a netbook in with Claude Code on it. my grandma is paying for my subscription (currently on the beginner plan but can probs convince her to let me go pro)I want to launch a B2B SaaS business (or multiples) and am able to dedicate my time to it 24/7 as I have rent, food and utilities taken care of for meAny ideas on what I should build? I have some sketched out but I'm keen to hear more
>>109082651a cock sucking machine that sucks your cock for a monthly subscription fee
>>109082651make a notes taking app, main usecase of vibecoding lol
>>109082651pics or gtfo
>>109082320what's the most efficient way to do Agentic coding while not spending billions tokens in the process?
noam shazeer heading to openaijeff dean probably got tired of listening to him complain about trannies and palestinians
>>109082681there isn't one
opus 4.8 on xtra high reasoning told me to reduce transatlantic latency from new york to london by connecting to canada instead of london (the server I'm connecting to is physically in london and this information was in-context)stuff like this doesn't happen often with modern LLMs, but it always makes me think if anything like this slipped past my attention at some point
>>109082531Wait 2 weeks and you’ll get a surprise from Grok aka Cursor
>>109082320have any of you managed to get jobs just off of vibe coding? i feel i have some pretty solid ideas that will be used/get attention on github but i'm uncertain if these will simply stay as memes or help me get into some sort of possible imployment
>>109082746most companies still want you to be able to demonstrate understanding of code, but "being able to leverage agentic programming techniques" is a skill employers actually take seriously now. not sure how much vibe coded repos help demonstrate that though.
>>109082746i'm currently looking for work and i have a nice little section for this stuff on my resume/portfolio. no one seems to care, yet... but i'm also not applying as a software dev (web dev/seo/marketing)there's been a lot of naysayers/gatekeeping regarding prompt engineers, but web devs are actually cooked. it's the most solved of anything
>>109082701default routing is not always optimalsometimes people get lower latencies by using vpns, which theoretically should not happen
>>109082746vibecoders are fucked in current market lol. There's too many software devs laid off by now and almost everyone know how to ai. Who will corps prefer people who had actual experience or some random losers who couldn't bother to learn to code?Ironically it was probably easier to get a job in industry pre ai with 0 experience.
>>109082876AI was only slightly hard to use in the very beginning, and the "left behind" narrative was always just cope. The tools get better and easier to use anyway, soon using AI will require basically zero skill, and the ghetto techniques people used in the early days are already starting to be counter productive.
>>109082920i know some people who still type shit like>you are le bill gates, come up with a plan...
Are there any local llms I could use for programming? Most of the ones I try have just crapped out nonsense.
my inference engine vs llama.cppthe model under test is qwen 3.5 0.8B (the tiniest one for faster iteration)
>>109082924qwen 3.6 27b is decentqwen 3.6 35b is not too bad if you need to offload (only 4b active)gemma 4 12b is also supposed to not be too horrible
>>109082920this. vibecoder as a job will never be sustainable and will always be a meme
>>109082681did it do some sort of test? in theory this isn't the dumbest thing ever - new york would in theory be a high congestion area, and technically halifax to london would be able to make up the 10ms~ from NYC > Halifax on the London > Halifax hop, the real question is if the vpn overhead would add too much latency for it to be worth it.
>cursor>antigravitywhy all these forks of vscode instead of just using vscode?
>>109082974probably don't want to rely on microsoft maintaining features they need
Claude keeps underestimating how long tests will take by a significant amount for depth search testing. It will think it's going to take an hour and 3.5 hours later it will check in and be like "yeah this is taking awhile lol guess ill wait some more"
>>109082974they've both stopped being forks of vscode - they're just codex style things now.editor boys stick with vscode or zed these days
made a little tracker so I can monitor the agents. right now have 70 running.workflow is generate 10 optimization ideas per kernel, implement them and run microbenchmarks, integration tests and merge the ones that work. using a gpu queue to avoid contention during benchmarks.gpt can be retarded to align a button on a webpage but then also manage an agent swarm and it seems to mostly just work with a bit of guidance.
>>109082974Cursor has had probably the best AI assisted autocompletion by far. Also one of its selling point was its code indexing feature. Claude code made both features irrelevant I guess but I miss the autocompletion sometimes.
>>109082058I don't get it, I'm autistic and a porn addict and never once have I thought "damn it would be great to be a woman"
>>109082320
>>109083100how much ram do you need for this?
Hey vibe coders!Who can optimize and modernize this old javascript timeline webpage?https://github.com/NUKnightLab/TimelineJS3Bonus points for making it local like this https://github.com/n8henrie/timelinejs3_local
>>109083185probably Codex
>>109083178Not that much. Each codex process consumes maybe half a GB. The usage is much more expensive. I went through the 5h limit for the 200 dollar plan in like 15 minutes.
>>109082320https://vcg.gitgud.site/
löl
>>109083236lul
>>109083236>as a fullstack engineerstop the cap
>>109083100you need a review layer because half of them will cheat and give you bogus numbers lol
>>109083236As a full stack engineer I must say that AI made frontend development bearable. No more CSS hell, just vibe code it.
>>109082320So how good is GLM-5.2 for coding?
>>109083280Haven't noticed a single instance of that happening so far. The main agent benchmarks the engine as a whole before and after merging all the changes that had positive integration tests and reports the changes to me in this format >>109082927I think gpt and LLMs in general tend to "cheat" in lying by omission to the user, not when the passing criteria is a simple arithmetic criteria, and there are multiple stages of gating and synthesis so a lie here or there wouldn't change the outcome anyway.
>>109083299>No more CSS hellI actually like writing vanilla CSS
>>109082446I've been trying to use DeepInfra's API since they're US based with a ZDR policy but their APIs cap responses at like 16k tokens which doesn't fit some of my use-cases. Still, their pricing is very reasonable though their speeds can be low with high concurrent use. With toolcalls the API response caps aren't a problem because very rarely will KimiK2.7 or GLM5.2 spend 16k tokens without using a tool(and tool use triggers another API request, resetting the cap) but if you need to pass it a large prompt that you need it to read deeply(which you wouldn't want to let it use tools for since tool availability makes models start lazy grepping) you can run into problems. This is probably a me problem though, most people would likely never run into this problem because most people want their models to use tools.
>>109082651I'm less interested in your vibecoding workflow and more interested in your anal stretching routine to be able to pull that off.
>>109082681Using multiple models. Big, capable model orchestrates while smaller and/or cheaper models implement. Cross model review of important decisions. Having Opus write a plan with you while GPT adversarially reviews it or vice versa actually saves tokens because you're less likely to have to go back and fix something later and one model can often spot efficiency gains that the other model left on the table.
>>109082839yeah route optimization as a service is a thing and can often shave a bit of latency off of naive routing. I used to use one of these providers for gaming (before vibecoding ate up all of my gaming time) and it would usually reduce my ping by about 25-30%.
>>109083331Really good, this latest generation of large open weight models genuinely feel comparable to american frontier models in a way that the previous generation never did.
>>109083460over which provider?last time I tried the z coding plan with 5.1 it was crapping out at 90k context for months with no response from the company which really put me off form using it ever again
so are we literally just not getting new model releases from openai and anthropic any more?
>>109083570no, it's over
>>109083570openai is still planning 5.6 though I doubt it will be held to the same level of scrutiny
>>109083520NTA but OpenCode Go has it.
>>109083902take it easy lil pup
5.6 bros...
>>109083520DeepInfra. Opencode has it too like >>109083634 said
>prediction markets say 55% chance we get fable back by julywtf this is lower than i expected id have thought closer to 85-90its so over
>Warning: You've used 50% of your weekly limit>Warning: You've used 51% of your weekly limit>Warning: You've used 52% of your weekly limit>Warning: You've used...
It emits (you) particles in thread watcher when someone quotes you in a thread.
>>109082746I already have a full-time day job in a non-tech field. It's how I pay for all this AI funtimes.
>>109084002Keep in mind the prediction markets already know when Fable is coming back because they simply asked Dario and he told them. They're all part of the same Tribe. So now Polymarket is keeping the odds neutral until more people get in, then they'll start really manipulating you to drop more cash on a bet they already know they outcome of. And around and around it goes.
>>109084172retard
>>109084197You sit there and tell us all that Polymarket and Kalshi aren't scamming and fixing. Go ahead.
>>109084002What's great is that if it's not resolved by the time a competitor has caught up, Anthropic will literally die.
>>109084172The volume on Polymarket is so low, it wouldn't really be worth it.
>>109082320Kek, two weeks ago I started bouncing an idea with LLM using the "Buffcat" meme and he seemed to go along but not really aware of the meme. Today I came back with a new question and in his thoughts there was:>Buffcat should do such and such>(This fits the Buffcat "chad" persona perfectly)He knows about him now.
>>109084295>heImagine sharing your most intimate ideas, hopes, and fears with another male. Couldn't be me.
>>109082378Infinite supply + zero demand = negative value (you have to pay people to use your slop)
>>109084310>sharing ideas with womenWomen are for baby-making, they have no brains to share ideas with.
>>109084331>If everyone has an oven in their home, there will be infinite supply of cooked food and nobody will pay to eat at restaurants anymore. They will have to pay people to eat their food!
>>109084336Right, you shouldn't be collaborating with some aggressive male who wants to dominate you and hijack your projects. You should have a female assistant who dumbly reflects your ideas back at you with some psychotic/BPD transformation so you can see what it looks like.
>>109084310>heI imagine this ai guy on the other side of the computer to be some old wise chinaman.
>>109084348why is chadcat eating snails
>>109084357Yes, I would rather talk to a man than a woman.
>>109084386
>>109084366>CHINGCHONGBINGBONG>CHINGCHONGBINGBONG>CHINGCHONGBINGBONG>solves your dev issue in 1483.1ms
>>109084381le chaton fat enjoying some escargot
>>109084420I wonder if there's a skill like Caveman, but like...>Claude say, good code need good chi.>Aligning with sunrise...>...>...>Feng shui optimized, auspicious day.>Ancestors proud.
>>109084381he's eating little snailcats
