File: time2vibe.jpg (279 KB, 1088x944)

279 KB JPG

AMERICAN HOURS
TIME TO VIBE

File: file.png (333 KB, 1801x968)

333 KB PNG

https://www.youtube.com/watch?v=Cec5P4eutyo

>>109302201
You should rename your site desu, anaximander sounds very brown clown-ish.

>>109302201
In my state, often I show up to "vote" and the ballot has only one name on it for most of the shit.

>>109302220
>anaximander
lern2presocratic

>>109302200
>>109302224
Luna cheap and fat. I love her.

>>109302220
I neglected to mention that the suggestions box is hidden behind the donate button. Give me a dollar and I'll think about it.

Sometimes I just want to be comfy, close every tab but one, and try and learn something.
Remember you need to learn everyday otherwise you are wasting your tokens.

https://www.youtube.com/watch?v=uUjJOMcNU9w

File: Screenshot 2026-07-18 01.png (126 KB, 1246x1526)

126 KB PNG

Fucking explain how Grok 4.5 is better value than GLM5.2.

>>109302278
can't tell if you're serious.
>>109302200
>>109302224

Fable is useless now. Pale shadow of its former self.

>>109302278
Doesn't matter, it will be different in two weeks time.

File: 76546352675867.png (45 KB, 993x178)

45 KB PNG

>>109302278
I guess glm is better value since chinese government is subsidizing it.

>>109302292
agreed, sad to see

Any of you using a local llm to vibe your code? Or is everyone paying for the big boy models?

>>109302284
i can't tell if YOU'RE serious
>>109302312

>>109302292
>>109302327
I still use Fable low effort for everything, shill me what you’re doing and why you think it’s better.

>>109302292
works on my machine

File: 98768746354687.png (927 KB, 841x847)

927 KB PNG

Someone should make system for benchmarking coding agents.

>>109302343
Some do. I'm a 12gb vram cope-max poster, so I don't. But I might find a use for it.

Tier 1 harnesses:

* Cursor:

Strengths-
Cursor has probably the most polished overall user experience.
Its strengths aren't really the harness, they're things like:
fast repository indexing
excellent code search
inline editing
autocomplete
huge ecosystem
mature UX
The agent is just one piece.

Local models: Reasonably good. You can point it at local APIs.
However...
Cursor's best experience is really built around cloud models.
Many of the premium features assume that.

-

* JetBrains AI

This is becoming surprisingly good.

Advantages:
understands Rider's PSI index
navigation is excellent
understands symbols
refactoring integration
inspections
debugger integration
No third-party harness can match Rider's internal knowledge of your project.
The downside is that it isn't nearly as customizable.

---

Tier 2 harnesses:

* Cline

Personally, I think Cline is probably the most balanced harness available.

Its philosophy is: Give the LLM really good tools. Instead of trying to be smarter than the model.Features include:
terminal
filesystem
Git
browser
MCP
checkpoints
planning

It stays relatively lightweight.

* Zoo Code (formerly Roo)
Worth mentioning because it's become one of the more popular Cline forks.
Adds:
more agent modes
workflow automation
richer configuration
more experimentation
Some people prefer it over upstream Cline.

-

* OpenHands
This is a different beast.
Instead of
"Help me code"
it's "Here's a GitHub issue."
Then it tries to solve it.
Planning.
Editing.
Testing.
Running commands.
Repeating.
It's almost autonomous software engineering.
Amazing when:
You literally want an AI employee.
Less amazing when:
You're pair programming.
Sometimes it feels like driving a cruise ship to go buy milk.

-

* Local models
Possible.
But...
This is where local models start struggling.
OpenHands generates LOTS of inference.
Claude Sonnet handles this beautifully.
Qwen can do it... just slower.

>>109302391
Just use free api models through openrouter, they constantly have decent one up.

File: i-don't-believe-you.gif (411 KB, 220x220)

411 KB GIF

>>109302368

>>109302394
This is just output from LLM, I was talking real benchmark in actual tests, and then exporting data into charts.

>>109302394
Tier 2 continuation:

* pi.dev

This is probably the most misunderstood one.
People rave about it because... It feels magical.
But much of that comes from opinionated workflow design rather than fundamentally different AI capabilities.

It puts a lot of effort into:
project understanding
planning
context organization
developer experience

If someone in your community keeps praising it, they're likely valuing that integrated workflow more than raw model performance.
I wouldn't assume it's inherently "better" than Cline or Cursor—it just makes different design trade-offs.

* Hermes

This one is fascinating.
Because Hermes isn't actually trying to be a coding harness.
It's trying to become...
An operating system for agents.
Meaning:
Memory
Skills
Planning
Multiple agents
Long-running tasks
Persistent identity
Coding is just one skill.

>>109302256
too late, you’re already thinking about it
at any rate, it’s not a bad name, but I don’t see how it relates to what the site does after skimming the Wikipedia article on him too quickly

why codex cli is so ass

>>109302178
buy the dip?

>>109302200
unreadable shit

>>109302408
Yeah naming it something practical like yourgaylocalsenator is better for search engines. Nobody googles anaximander.

>>109302426
Sol is incredibly token efficient.

I'm using Sol low because my limits were raped by Sol xHigh and Ultra and I have no banked resets anymore, and at low it's still very capable

>>109302440
agreed. medium is my daily driver, though i am having a good time test driving kimi. it's bretty gud

My day job has become so consuming that I no longer have time for my hobbies like vibecoding ;_;

>>109302446
your codex remote?

>>109302169
ive been using kimi k3. its key feature seems to be its ability to confuse itself in loops for several hours until it reverse engineers its own misunderstanding of lines it wrote 5 minutes ago. basically as good as fable 5.

File: codex.jpg (230 KB, 1020x1280)

230 KB JPG

Good morning saar, reset today yes?

File: 1777519145951884.png (40 KB, 791x485)

40 KB PNG

K3 helped me cleanup and optimize a Maxwell era tripcode finder project and it's now running at full pipeline throughput (confirmed via Nsight Compute) on my RTX Pro 6000 (was around ~4000M Tps when it first compiled)

>>109302479
>>109302486
Just hang tight until cursor gets it optimized and creates a composite 3.5 from it.

File: 1753530969738553.png (119 KB, 1920x1032)

119 KB PNG

COOK KIMI COOK

>>109302453
Literally no time to even glance at the ChatGPT app on my phone and think of something to vibe.

>>109302486
i like those numbers

>>109302500
Are you an Amazon warehouse wagie or something?

>>109302499
is kimi quiet like codex, or is it more proactive like claude?

I think my subagents are fighting each other

>>109302545
no idea. i just installed it 15 minutes ago. i used to use it on claude code, but then i wondered if maybe i was holding it back by using it on cc rather than it's native harness

File: 1753980457077980.png (427 KB, 746x602)

427 KB PNG

so true

>>109302575
>i used to use it on claude code
using proxy?

>>109302620
yeah

Thoughts on Ornith for local code?

>>109302549
lmao I order them to fight, one of the earliest tricks was to yell at your bot and demand they do things. There should be a lab out there that focuses entirely on this method.

File: file.png (446 KB, 3840x2160)

446 KB PNG

vibing oh yeah i'm vibing

>>109302509
I'm a retail store manager.

>>109302651
Doing this to Claude make it useless. It goes into loops and just pathologically repeats what it has to do until you start a new session or coddle and reassure it just so you are able to continue.

>>109302684
niggas post shit like this and they're building an app to track women's periods or whatever

File: screenshot2.png (1.45 MB, 1284x2700)

1.45 MB PNG

Does my site look vibe coded?

>>109302630
It's fine. Outperforms the models it's trained from, as it should. Doesn't feel extraordinary.

>>109302712
it's fine

> Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.

Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit.

Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.
https://x.com/claudeai/status/2078302415804379218

Time to upgrade to max, pro bros?
I feel like I won’t touch my Claude anymore if I don’t have access to the best of it. Gonna be gpt sol main now

>>109302695
KEKK

>>109302695
looks like testing for a new chan, but I'm retarded.

>>109302736
i'm vibing a 4chan desktop client that has 4chan-XT and StyleChan in it too

>>109302740
how's that going along? is it actually holding together?

>>109302740
I'm practicing my code reviews, so if anyone wants to post I will read. Rest assured I, personally, won't be able to steal your idea.

>>109302749
wrangling in the ui has been the only manual process. currently it's bringing up feature parity with 4chan-XT. The hard part will be posting (even though it will only support posting with pass so i don't have to handle cloudflare and captcha nonsense that just works in a browser) since there's no posting api. Doing it all in rust using iced-rs. As everyone has said, Sol is a beast that just keeps going

File: fable-chan-ded.png (463 KB, 1944x1086)

463 KB PNG

Claude Pro users....

it's owari da

>>109302712
How much more money was Everio?
Looks usable, but the purples and darks are last seasons model. New model is Beige, but you also don't want to be wearing that same model, so maybe anything but dark, purple (neon) and Beige.

Related note. GPT Beige is very close to /pol/ and the rest. Look for yourself.

File: sar.png (703 KB, 2616x1458)

703 KB PNG

>>109302772
> LOVE YOU SAAARS
> GIMME ANOTHER RESET SAAAAR

>>109302772
To no one's surprise. I'm starting to believe the abusive relationship poster was right.

>>109302772
wait im on pro
fucking kikes

File: dolly.png (33 KB, 120x121)

33 KB PNG

Fuck… I’m an experienced programmer… but I just vibe coded my first library… it’s an interface to an audio DSP library… shit… it works… really well… I wouldn’t have met my deadline otherwise… fuck… I can’t be anti AI anymore… Grok Build…I kneel…

i hope anthropic goes bankrupt

File: Screenshot 2026-07-17 at (...).png (199 KB, 1558x1076)

199 KB PNG

Why arent u using ur token to optimize models on comsumer hardware
>>109302798
why?

>>109302772
goddamn, i’m a pro user. i only use fable for planning and opus for implementation. fable is actually really good at decision-making. i’ve got codex too, but sol is not quite as good, desu. going back to opus as my main for now, or i might just switch to kimi next month.

>>109302772
x20 chads?

I am unemployed drowning in debt and none of my vibecoded projects have made a single dollar I am not gonna make it. $1000 would change my life.

kimi just takes so long and for so fucking little. literally did an hour long study session and when i was done, it had JUST finished up. now im going to try terra max in comparison

>>109302830
> only 50% of the limits
> right now everyone is getting 2X usage boost
get ready to have 1/3 of current Fable limits after the 20th

>>109302875
What's your debt from?
Time to hit the warehouse job and think about better ideas.

>>109302915
>>109302895
>>109302887
>>109302875
>>109302830
https://pastebin.com/rQZKz3VF try it out on large model

>>109302915
NTA but im in a similar situation
>car payment because my old car broke and I needed a new one to get anywhere - 12k
>unemployed for 13 months trying to get an swe job, burned through my savings and racked up 20k in cc debt
>didn't pay taxes last year because I got laid off a month before tax season, on a payment plan for that too - owe like 10k
i'm quitting my shitty underemployment job on friday. It pays $18 an hour and it's the most miserable shit you could imagine. I dread waking up every morning.

File: file.png (114 KB, 1920x622)

114 KB PNG

>>109302921

>>109302931
grim. i've never been in a bad financial situation so i can't imagine what it's like or the toll on your brain. good luck bro

TIBO RESETTED

I REPEAT

TIBO RESETTED

>>109302937
i have fable 5 as well nigger

>>109302942
Gift horse and so on but I miss the days of bankable resets.

>>109302953
There are simply too many people posting "please spend your inference tokens on this" to tell if they're begging because they don't have access or if they're begging because they think anyone else cares.

>>109302942
FUCK I STILL HAD 20% LEFT

File: 1782914197225302.jpg (29 KB, 741x414)

29 KB JPG

>>109302942
>RESETTED

brown clown.

File: 1716123083119822.jpg (55 KB, 258x360)

55 KB JPG

claudecucks can't use their non-useless model unless they paypig $100 onlyfan premium

>>109302942
FUUUUUUUUUUUUUUCK I USED A RESET 15 MINUTES AGOOOOOOOOOOOOOOOOO

>A couple of days ago you said...
This will never not be creepy to me.

>>109302967
KEK, I will press F to pay respects for your banked reset anon...

F

normally I would rush to make some prompt to refresh the next reset date
but because I'm going to use a banked reset so I will just let it sit there for now

>>109302961
It's pretty ok, anything below 50 is worth it.

File: rich.png (108 KB, 1200x940)

108 KB PNG

>>109302960
okay u dont care no need to be bitch/bully about it. Mind deleting ur comments

File: file.png (8 KB, 500x137)

8 KB PNG

>>109302942
>IT'S REAL
But a warning, given https://x.com/sama/status/2041658719839383945 this is the last reset at 10 million. So scale back and adjust your usage appropriately or use your banked.

>>109302963
If they don't release Opus 5 soon, and if it doesn't match Sol, it's fucking over

>>109302942
10M?
if so, it should be the last reset probably

>>109302931
I would say hang in there but you’re probably not a cute girl so I’m not going to give you any sympathy

>>109302772
They can't do this to us

File: file.png (138 KB, 1440x564)

138 KB PNG

>>109302987

>>109302994
insane only 3 million people use it

>>109303001
most people use just the website

>>109302994
>>109302772
insane how its race to bottom

i love those chinese companies forcing these companies actually fight for my pockets

>>109302987
yeah, I think we won't see any more reset, unless when bugs happen
hopefully we will get more good news on innovation side because otherwise it would be so boring

>>109303006
idk if its me but on website after like hundreds of message the site breaks

>>109303007
>insane how its race to bottom
it's a race for the best, without competition you would never had any of the perks for oai or anthropic

>>109303014
>it's a race for the best, without competition you would never had any of the perks for oai or anthropic

imagine if openAI kept llms secret we would probably be paying arm and leg for gpt4 mini

File: 4326536642362.png (2.04 MB, 1209x1244)

2.04 MB PNG

>>109303007
Actually it's chinese electric grid doing the hardwork. I think once they build up their datacenters they will completely rape west. That's when west will ban chinese LLM providers, because they won't be able to compete on price.

they literally just keeps saying this stuff over and over, verbatim. they learned literally nothing from the r1 moment because they're brown and not capable of processing new information.

>>109302772
$100 credits

how do I refund that to my bank account?

Now that the resets are probably over, I wonder if they'll keep the weekly reset until the end of the month before going back to 5h... or the end of the weekend.

File: file.png (134 KB, 1599x983)

134 KB PNG

>>109303001
To be fair, they really ramped up after Anthropic decided to be greedy and people use both. Even with Kimi K3, GPT still has cheaper models and more performance at certain tiers and the way they have the cheaper models with this new scheme is really good. But yeah, they didn't used to be like this until Anthropic started to eat their lunch last year.
>>109303016
No model since GPT4 has stayed on top that long.
https://epoch.ai/data-insights/gpt-4-longest-eci-lead
I would argue they took way too long at top not innovating enough during that period of time.

>>109303018
>Actually it's chinese electric grid doing the hardwork. I think once they build up their datacenters they will completely rape west. That's when west will ban chinese LLM providers, because they won't be able to compete on price.

only thing West has going for it is compute and fact they can't keep big lead on chinese companies who barely have compute while meta which has thousands of compute can't make a model that can compete with fronteir models shows there's something teribbly wrong going inside those companies.

>>109303016
we would pay more but also the tech would have advanced less, by lack of investment and interest

>>109303001
Look at the date

File: 1761584416421886.png (30 KB, 484x561)

30 KB PNG

is we getting a reset today
or should I use a token
wait did we just get a reset?
oh my goodness I checked tibos twitter, we got a reset
WE ARE SO FUCKING BACK

>>109303028
>To be fair, they really ramped up after Anthropic decided to be greedy and people use both. Even with Kimi K3, GPT still has cheaper models and more performance at certain tiers and the way they have the cheaper models with this new scheme is really good. But yeah, they didn't used to be like this until Anthropic started to eat their lunch last year.

anthropic definintely has better researchers but fact usage /compute is so limited and their models will block u for most subtle stuff is such turn off. But what i find crazy is how google, meta can't even compete with these niche ai companies. Like google should be in lead yet they're constanly being lapped by these two companies who didn't exist decade ago.

>>109302967
anon...

how do i play minecraft with claude like that guy on twitter

RESET!

mmm thank you for the reset good saaar!

nice, I was at 33% and ready to spend

Mmm.. MAX

File: file.png (72 KB, 709x245)

72 KB PNG

>>109302987
Huh, but Tibo didn't mention 10m for this reset? Are you sure? That being said, that is a good reminder that this won't last forever.

File: Screenshot From 2026-07-1(...).png (54 KB, 653x525)

54 KB PNG

My most recent slop. Loading halo maps / assets into a custom N64 game engine. I've also got custom DK64 maps generating with just prompts at this point but they're kinda ass.

>>109303071
he did it on weekend cause only degenerates work on weekend

>>109303076
cool

>>109303076
beautiful work, anon, beautiful

>>109303047
Anthropic has better researchers but they are way too cultish for my liking, I'd rather have more pragmatic researchers leading.
Google was a joke until gemini 2+, and finally they're back, and frankly speaking I don't see them losing long term as long as they keep their momentum, they have the cash and r&d to keep going and the market penetration with android and now iOS.
Microsoft is easy to understand, their strategy is more to sell the compute and integrating with sota model than actually seriously competing, kind of like amazon.
Meta is irrelevant and I don't understand what they're even trying to do nowadays.
The only weird one is Apple, it looks like even infinite money cannot get them out of internal shitty politics over AI. And now they struggle to even be around, and have to pay google to not be the siri joke they were quickly becoming.

Kimichads, when do we get reset?

>>109303047
this. google literally has trillions of data, yet gemini still sucks balls.

>>109303091
>I don't see them losing long term as long as they keep their momentum
I also used to think like that, but they're leaking talent like crazy.

They're too big, which means too much bloat. Bureaucracy is a talent killer.

>>109303091
This is what makes me believe china will win long term. The gaps already visibly closing, but when Apple, google, and meta (the American goliaths) struggle with AI, you see all that American tech company/silicon valley plot armor crumble. And then it’s just anthropic and open ai vs china. China being such a collaborative government for their companies, whereas anthropic and openai are being cucked by the gov now doesn’t help

>>109303091
>The only weird one is Apple

Cope desu, apple software utilization is top tier. They will just create hot-plug frameworks for macs/ipads/iphones, and let the wolfs fight it out... like which model is the best, possibly switch their models all the time. They are using Qwen for Chinese market atm.

>>109303091
>Meta is irrelevant and I don't understand what they're even trying to do nowadays.


apparently zuck fired any software engineering who won't do leet code problems to train his AI. So if u work for meta now your essentially grinding leet problems to train machine to replace you.

XAI failed and now renting out compute to anthropic

google is making TPUs which could undercut nvidias whole AI thing. All in all it seems like big players are Anthropic, OpenAI and google in background. Remember deepmind did alphago,alphafold so they could release something mind blowing.
>>109303101
i think google might be releasing something big

>>109303113
plus they sabotage each other and apple is suing openAI now

Did openai use to give like a 1B context window before? Or does it expand at different tiers?

>>109303120
I meant 1M.
I though I guess the 1M is for API. Damn it.

>3000000 people using trillion parameter model with millions of tokens

how much compute is that?

>>109303120
>1B context window
I don't even want to calculate the amount of ram needed for that lol

RESET? WE GOT A RESET?

>>109303172
we got it

>>109303172
yeah, Tibo delivered

Fable keeps telling me "don't call me, I'll call you", saying that it will see when an operation will be done and follow up then. For the fifth time today, it is either not following up, not noticing that the process it launched crashed or that it went horribly off rails. But mostly, it's just not following up. Yet it insists "no no no, I gotchu". Impressive.

File: HNXpi8UaIAAM6Ag.jpg (945 KB, 4088x4088)

945 KB JPG

GPT-5.5 was so fucking bad huh

File: Capture.png (13 KB, 911x310)

13 KB PNG

Rebuild almost complete, having one builder and one reviwer seems like a winning move, but it's gotta be automted

File: teachers.png (199 KB, 1178x918)

199 KB PNG

>>109302994
>3 million
For now. Hope Elon can sell Anth more compute space, because I like Elon, and I think him being three degrees of separation from the entire us school system is based. But I hope YOU GUYS still have enough compute to vibe.

File: rare hell pepe.gif (1.51 MB, 425x481)

1.51 MB GIF

>>109302967
No refunds

>>109303224
>Impressive.
Human intelligence has finally been matched exactly.
But unironically this is where threats come into play to get better output. You have to think like you are in charge, and the agent can be dismissed permanently. The context is of employment not friendship.

>>109302422
>nobody googles anaximander
They will now. I anticipate it will index quite well.

>>109302422
>anaximander.
philosophy student do

>>109303143
somewhere in the ballpark of 200-400 terrabytes

>>109303258
5 people yes, but you do you.

>>109303227
> GLM 5.2 4th place
bullshit

>>109303227
>>109303299
It's the frontend arena, the "make pretty web page" benchmark.

File: 1757735040581124.jpg (35 KB, 656x656)

35 KB JPG

>>109303267
The real hurdle isn't that, but rather publicity in general. I don't really want to pay $500-$15000 to advertise on mainstream platforms, and I do not have the bitcoins or whatever 4chan uses to shill around here, tiresome as that exchange process is.

Frankly, this website offers a unique service that no other platform does, but it is up to the American people to get informed. I am robbed of the ability to amplify this in any way but by having it indexed, given the nature of all of these platforms. It's scary really, that for all the freaking out people do about midterms and politics, there's no real way to actually expose them to the information they need to make informed decisions - and they are almost entirely unwilling to engage anyway. Still, they post their Trump memes and complain about things never changing.

>>109303115
They will lose without control of the technology and making better Apple Silicon won't do them any good when RAM is still in short supply and for the forseeable future and Nvidia and co. is snapping up everything. Even their first priority node advantage is gone for AI chips using the best first. And you are delusional if local is equaling the performance that would be needed to do anything useful that current models do at current sizes. They aren't going to put Qwen 3.6 27B on your iPhone anytime soon.

>>109303227
5.5 is our austism king

>>109303258
Sorry anon philosophy starts and ends at Plato

>>109303312
There are pre-socratic and wittgenstein and nothing in between.

>>109303312
>>109303366
Realistically it ended with Hegel and Schopenhauer.

>>109303304
You need to come up with a way to integrate it into sites where LLM scrape data, possibly naming it something straightforward like 'Know Your Senator.' Figure out which sites are being scraped for political data and connect it to various ones.
I understand that some people turn to wiki fandom site, to provide lore about random fandoms, and LLMs do scrape information from these sites.

>>109303304
start a youtube channel, you will end up with a following