A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.You use Git — right, anon?## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## News (both past and future)- 2026-09-29 — OpenAI releases GPT-6.1 Sol- 2026-09-28 — Anthropic releases Sonnet 5.5- 2026-09-22 — OpenAI releases GPT-6 Sol and Luna- 2026-09-22 — Anthropic releases Opus 5.5- 2026-09-22 — Anthropic increases subscription plans's 5-hour limits by 20%- 2026-09-14 — Anthropic reduces subscription plans's weekly limits by 17%- 2026-09-12 — Anthropic suggests to pace the frontier. OpenAI agrees in principle.- 2026-09-10 — OpenAI pauses new sign-ups for their $200 subscription- 2026-09-04 — OpenAI releases Astra- 2026-09-01 — Anthropic releases Fable 5.1## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-code — probably generally better currentlyhttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/cli — no 5h limit for only $30/month## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favoritehttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109935982
>OpenAI
>>109940099>>109940129>>109940150>>109940151>We've halved your allotted compute on your """20x""" plan so more important people can pay us more for the same product's outputs in less time!i guess i'm moving to claude
>>109940179>Anthropic isn't going to copy this at the first opportunity
>>109940197well they haven't yet and opus 5.5 is sota so i'll enjoy it while it lasts i guess
>>109940179Will do it too, 6 more days before the end of my $200 plan. Time to waste those banked reset.When I moved to the $20 on chatgpt, they nerfed it one week after, then I moved to the $200 the next month, and then they nerfed it 3 weeks after. Enough is enough.
>>109940223inb4 you're the reason they keep nerfing plans and you bring it to claude too
>OpenAI wants to be the new Microsoft Officeyawn
>>109940237Well, if Claude start to nerf shit after next week then....
>>109940179subscriptions are only good for getting your feet wet. once you do you should spend time replacing subscription reasoning with your own model stack. once you do that you start actually learning how models work and why subscriptions are to be used maybe 20% of the token time. most serious teams shoot for 5% "frontier". so your $20 or $200 should 10x or 20x usage because you are learning to build independently.
>shameless Apple mimickingpathetic
>>109940257okay well i'm still going to buy a claude 20x sub because i get more out of that
HOLY SHIT. THIS CHANGES EVERYTHING (nothing actually changes)
>OMG LE HECKIN' TIBO RESETyeah my weekly allowance is still halved though
Wow must be a lot of resets coming, it's taking awhile to load.
well that was shit
>>109940291that's fine if this is all you know how to do and don't want to learn anything else. it's more than enough for a lot of people. I push billions of tokens a week so this kinda thing doesn't work for me.
was that even 20 products or did i fall asleep and miss like 16 of them
>>109940378>6.1 Sol>dots>Jev with Visionthere was a 4th?
>>109940378anon.... it was you all along....you are the product
Again the 100 math results were mentioned, but not released.
I use planning mode a lot and generate a phase-oriented outline. That's where I've found value in Claude Code's quickness. I'm okay with the actual coding and implementation part being slow.I have a couple of laptops with i5-8250U CPUs and around 8 to 16 GB of RAM. What options do I have for running something good locally?Make no mistakes.
>>109940428even people with good specs don't have good choices for local models.
>>109940428>8 to 16 GB of RAM>running something good locallyIf those arent on some lite linux distro forget it. You can get summarizers and document sorters with the small gemma 4 models. but anything else will be super slow or not fit. maybe a small qwen for some coding but if its not the 27b 3.8 its not great either. I guess depending on how many laptops you could set up a swarm of small models for fun? but you are going to get like 10-15 tk/s max on just ram.
>>109940428use claude to build automations designed to replace claude
>>109940450Damn, I’ll keep searching, but for now, $20 for Claude Code is a deal that pays for itself. I doubt it’s going to stay this cheap, though.>>109940479At the moment, I use Gemma-4-E4B-it-Q4_K_M and OpenHermes-2.5-Mistral-7B.Q5_K_M with llama.cpp. Integration is my biggest pain point right now. Those models are okay, but I don’t trust them 100%.
My personal benchmark is for every new model release I try to get a model that manages to make a chrome extension which gets rid of pic related (and I don't mean the popup, this is trivial, I mean the 5-15 seconds of fake buffering that they are doing before the video starts)and so far none of them have managed, including Fable and Astra on Ultra.AGI has not been achieved
>>109940501>I’ll keep searching >>109940501>>109940450doesn't know how to use models or computing in general
>>109940549why would they fake the buffering?unless your extension plugs into the nearest server, you're not gonna be able to fix that
>>109940315Maybe I can swap from sol medium to sol highWait a minute it's 1/5th the price of astra, not 1/5th the price of sol. I thought I was getting a 5x efficiency upgrade.What a disappointment. Well fuck this I'm going to bed, not waiting around for that trash. Probably not even going to use it since it'll still blow through my usage.
>>109940549>fake bufferingkek brainworms
>>109940577anti-adblock measurenot him but I can't even play 720p videos anymore, they buffer every 2 minutes
>>109940542You can try the qwen 3.5 9b and bonsai 2 but they arent to be really trusted either. its okay if you let them reason but it will take forever at those speeds.If you are willing to rig you can take a 1070ti which is like $90 and hook it up to run a qwen 35a3b which is a lot better but you will have to set up a egpu and the loading the model times will be horrible and i dont know if it will be actually fast as it loads and unloads experts could drag you to single digit tk/s or prefill.
>>109940583Oh it's the same api price as gpt 6 sol. I'll switch to it and stay on medium then and give it a shot.
At work so not watching but was the alleged reset banked or normal reset, deciding if I need to use a banked to keep working as I was expecting a normal one today
>>109940577>>109940588it's a literal thing. I have gigabit fiber. 4k video streams on netflix load instantly but youtube takes its sweet time buffering 20 seconds before loading a 720p videosee https://community.brave.app/t/how-to-deal-with-fake-buffering-on-youtube/654516
>>109940626banked
>>109940628>i have niggabitfiburr>but i use brave>why is it not working??? kek wormed
>>109940628>bravengmi
>>109940633can you even read you turbofaggoti am using chrome, the brave forum link is just an explanation of the issue
>>109940610thanks dude I will explore the 1070ti option futher.
>>109940628mine does it for maybe 2 secondsstill don't see how another extension would fix this - if it's fixable. would probably have to fix ublock
>>109940645google fingerprinted you as a brave user, uninstall brave
>>109940645>chrome
>>109940645Doesn't change the fact you put in a brave link
>>109940549which tools and mcps are you giving the model to work with?
Is there a way to update Claude's context without forcing a response? For example I want to tell it I changed color from one to another and to remember that in the future.
While local models are the endgame for this AI bubble, there’s no reason not to abuse subsidized tokens as much as possible in the meantime. When the bubble pops, hardware prices will come back down, and you can run the local larger models you want.
the only good thing that happened today
>>109940702sounds like you're looking for the pre-prompt? or just have it update memory
>>109940702auto-memory should do that all by itselfhttps://claude.dev/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models/#then-memory-in-claudemd-files
>>109940549There is a trick for that: load comments, if you scroll long enough for the comments and the right side bar, you get the video buffering less and almost insta load.Try it you will see.
>>109940718>When the bubble pops
>>109940718>here’s no reason not to abuse subsidized tokens as much as possible in the meantime.yeah using it to make your set ups better.>the bubble pops.If. we could still be on track for own nothing and renting compute they will try it.
25% more usage for 250% the cost... I am genuinely speechless.
>>109940749did you miss the sol price-cuts? sol costs now half as much. opus-class model for the price of sonnet.
>super long days at work this week>don't have time to vibe until friday
Newfag to vibecoding stuff.I wanna restart my Claude instance and get back into working while not eating so many tokens. Should I tell it to write a design document based on what it did by now? Anything else worth throwing in?
>>109940762no, don't restart but keep using your warm cache
>>109940775cache stays hot across instances dingus
apparently 6.1 sol is monster
>>109940789more like a gobblin'
>>109940762don't listen to dumb retard over here >>109940775>Should I tell it to write a design document based on what it did by now?yes. Or, if you're project isn't that big, just tell it to update memory. When you boot up the next one, tell it to check memory (the save it under your current user directory)in my experience it's cheaper to kill them once their task is done to save on context than it is to care about cache (unless you're only using 10-20% of context)
someone let me know if it's over or we are back
>>109940799>the save itthey save it*
How good is 6.1 Sol?
>>109940808it so over AI winter is here. expect nothing but gimmicks to get more non coders to sub.
use case for dots? Cheating in video games?
what IDE can I use on a 4gb ram laptop without it crashing intellij-idea is too bloated and makes it froze
>>109940829>4gb ramyeah can you even run cargo or cmake with that
>>109940799It's eating 10% of my usage from just sending a new message so I imagine it's time?Or is it time for me to stop asking it within Claude website and install the Code malware?
>>109940841>It's eating 10% of my usage from just sending a new message so I imagine it's time?it's fuckin time anon>Or is it time for me to stop asking it within Claude website it's fuckin time
seems OpenAI is falling behind Anthropic, we need to stop the frontier
>>109940789It devoured sol 6 pretty fast.
>>109940839I never tried cmake I am not used to c only java
>>109940819it's not
>>109940787restart still leads to cache misseshttps://code.claude.com/docs/en/prompt-caching#actions-that-keep-the-cache>Actions that keep the cache>These actions either append to the end of the conversation or don’t touch the request at all. Some of them, such as editing CLAUDE.md, keep the cache for the same reason the change doesn’t reach the running session until /clear, /compact, or a restart.>Editing files in your repository>Editing CLAUDE.md mid-session>Changing permission mode>Changing output style>Invoking skills and commands>Running /recap>Rewinding the conversation>Spawning a subagent
>>109940857i don't think you have enough ram to do your tests and builds, get a better rig thirdie
>>109940864>restartkeke is this a claude thing? i'm used to the way agy works and what i was informed by gemini
er was no reset even announced? We got pointless cloud shit, some weird always on assistant (I don't want this) and my usage is going down by half on astra.
>dot.com redirects to grok bot
>>109940884tibo said current $200chuds get extra token credits lol
>>109940884>>109940893I have 1 banked reset come in
>>109940893i think ill let them keep that
is astra worth it?
can someone seriously give me a use case for dots? I can't work out why I would care, is it for middle managers or something
>>109940799don't manually manage your context. if you want to save money do as codex does and set a smaller context windowhttps://x.com/thsottiaux/status/2076543065045795309>[...]>The actual reason [for a lower default context] is the what you can see depicted in the chart below, which is the difference in the orange line and the blue line. It is caused by overall cost of cache reads going up with the size of the context being shuffled back and forth between toolcalls. The sweet spot in terms of cost is therefore not necessarily to use the maximum possible context length.>[...]
>>109940917competing with muse spark. life organization. not sure if it does nsfw like muse. i'm happy with my spark, it's not a genius but it's a good worker bunny with access to github and a bunch of normalnog apps i don't use.
NIJIKA IS CUTE!!!!
>>109940917no, I watched their ad for it and it's more "choosing where you can eat!!!!" stuff lol I don't think they know what to use it for either
>>109940944>I don't think they know what to use it for eitherbecause it's an afterthought competitive product. they couldn't let zuck capitalize on an empty niche.
>>109940929it's just called muse.https://muse.ai/muse-spark is the name of a model.https://dev.meta.ai/models/muse-spark
When I get home, gonna try Ponytail
>>109940919I never compact, I always kill them at 20-25% (unless the task takes them over, then after task is done), and always at full 1m windowsome random xitterfag won't convince me of going against what 3-4 years experience with llm's taught me
>>109940950>it's just called 4chan>/g/ is the name of a board kek>>109940955>poonytail>not in OP anymorekek absolute shit
>>109940956>tibo>head of codex>random xitterfagfirst day here?
>>109940917meta muse competitorbut normgroids already get muse for free. i don't know who is paying for this and i don't know why they thought it was a reasonable tradeoff for mangling the $200 plan.
Opus 5.5 spent his 5 hour usage searching every research paper on the web yet again.Not that I can complain since I keep asking for extreme scientific accuracy so I will just have to keep going at it slowly I guess.
>>109940970>normgroids already get muse for freeit has a full ubuntu vm with 99gb free. for free. wtf. too bad it's too late to mine on that bitch.
>>109940965bad metaphor.muse agent will soon be powered by watermelon, not muse-spark
>>109940969he is to meI don't care about your gay resetsand believing anything they say is a sure sign of retardation
...hello?
>>109940988>obsessed with watermelonwhat's that beeping noise? anyone know? when the model updates to muse friedchicken i will call it muse friedchicken
>>109940990tibo literally said in the same tweethttps://x.com/thsottiaux/status/2076543065045795309>[...] The benefits of higher context lengths are mostly overall speed (as you don't wait for compaction), ability to deal with humongously large input and potentially cost if the system is well tuned and you hit your cache perfectly. [...]
>>109940315>(nothing actually changes)the cache cost is actually huge
>>109941003your language lacks precisionsa model doesn't make a personal agent.muse-spark is just a model and also used in meta.ai, muse code, ...
>>109940965everyone who hates Ponytail worships demons.
>>109941004tibo hits the reset button like a good monkey. don't take his tweets too seriously.
>>109941004so fucking what? go suck his cock>the benefits of higher context lengths are mostly overall speedno it's fucking notit's that the closer it gets to filling up it's window, the more retarded and unpredictable it gets
>>109941016>your language lacks precisionskek maybe it does, that's why i vibecooode. the agenda is to fully merge the ecosystem and put the personal agent in the harness eventually, it seems.
>>109941001i don't even bother with openai releases in the first 2 daysanthropic manages to get their new models out on day 1openai always takes at least a day for me to drop them in codex
>>109941030>go suck his cockhe's kinda cute desuand he has a LOT of tokens...
>>109941030>it's that the closer it gets to filling up it's window, the more retarded and unpredictable it getsthis is not true anymore for modern models. neither anthropic nor openai have that problem.
>>109941062not my experienceit's not as bad as in the past but the less garbage in the context the better the outcomes tend to be
>>109941062it's might not be as bad as before because the overall baseline has gotten higherI see no reason why it shouldn't still hold true
>>109941001The order is usually>US Pro users>EU/Asia Pro users>rest of the world Pro users>US Plus users>EU/Asia Plus users>rest of the world Plus users
>pre-nerf opus 5.5 with good limits>not-sol sol 6.1 with decent limits due to 50% cache cost dropyea I'm thinking codeslop time, things probably will get worse pre-ipo rather than better
>>109941022works for glue code, but absolutely shit for my usecase, which is dsp code, topologies, and rust
>pre-nerf op--fuck off
