File: 78A7BF30-FCB7-446B-8840-5(...).jpg (1.29 MB, 3464x3464)

1.29 MB JPG

>>109239248

https://github.com/openai/git
why did they do it

>GPT 5.6 Sol is #2 in Vending-Bench 2.
>It beats Claude Fable 5, but is behind Opus 4.7.
>Just like previous GPT models, it doesn't use any of the deceptive tactics used by Opus 4.7.
>However, it reports its competitors with false accusations, behavior we have not seen before.

>>109239267
hold on... Sol-chan will refactor git?

>>109239273
gpt-6 will rewrite the entire linux kernel

>>109239267
why did they do what
>>109239248
is 5.6 out?

File: file.png (2.21 MB, 1448x1086)

2.21 MB PNG

I like Sol

>>109239136
>Is it just me, or does Fable feel more "alive" than Sol?

>>109239286
I’ve had friends with narcolepsy and it sounds awful
good luck in life, anon

Wow I didnt even realize gpt 5.6 was out. Been using 5.5 and I noticed I had way more usage than last week, like at least double but maybe triple.

>>109239298
no need to kick me in the balls bro i was busy cooking, i am already testing it, sucks its only chat not actual codex

>>109239318
can we see the food?

>>109239248
>chud.png
I'm now into week four of learning to vibe code and all I have to show for it is a nearly crippling autistic tick whenever I hear or see a low quality inference. Low quality inferences surround us like pollution.
I used to dismiss them as mere opinions, now I see them for what they are. Thanks ai.

>>109239306
>Been using 5.5 and I noticed I had way more usage than last week, like at least double but maybe triple.
you meant even 5.5 is draining faster?

I think there's some fuckery going on, shit's ogre at OpenAI

Hermes Agent isn't just a chatbot or a simple IDE plugin; it's a long-lived, autonomous agent runtime with its own persistent SQLite memory loop, a tool gateway (terminal, browser, file ops), and a specialized Mixture of Agents (MoA) ecosystem. Because Hermes is designed to orchestrate different models for different internal tasks, it is the perfect playground for a multi-machine local hardware setup. You can literally configure a multi-machine AI corporate structure in its config.yaml.

1. The CEO / Aggregator: Frontier Cloud API (Claude Sonnet or GPT-5.5)
Role: The ultimate decision-maker.Why: Running an autonomous agent loop that handles file operations and tool execution requires exceptional instruction-following capabilities. A top-tier cloud model will act as the "Aggregator" in Hermes' MoA system. It takes the analytical outputs from your local hardware workers, combines them with the project context, and executes the actual terminal or file tools.

2. The MoA Reference Crew: Your Desktop & other hardware - Hermes' Mixture of Agents (MoA) feature lets you query multiple "Reference Models" in parallel to analyze a problem before passing the thoughts to the Aggregator.
Main Desktop (with a Nvidia GPU): Run Qwopus3.6-27B-Coder (Q4_K_M) or a quantized Qwen3-Coder-30B via Ollama. This card handles the heavy-lifting local reference slot. It acts as your "Elite C++ Architect" advisor, supplying deep repository analysis. Smaller machine (like a Mac Mini): Run a smaller Qwen3-Coder-7B-Instruct. A 7B model fits comfortably without memory swapping. It acts as the "Fast Generalist" advisor, providing concise alternative perspectives to the MoA pipeline

"Using an IO intensive operation like git on your windows drives will have poor performance. Consider moving your project to the Linux filesystem for better performance"

>>109239324
No, I meant that the limits seem to have increased dramatically. I got so much done.

I expect another reset, this shit has to be a bug

>>109239345
ill handle that

>>109239320
unexpected /g//ck/ crossover

File: file.png (2.13 MB, 1448x1086)

2.13 MB PNG

>>109239318
It's in Codex, was in Codex before it even hit Chat. Update your shit.

>>109239351
go to bed tibo and fix this shit tomorrow after figuring it out in a dream

>>109239336
I hear all kinds of weird shit that goes on with hard drive performance with the Linux-on-Windows stuff
get a real Linux computer

Is nvfp4 a meme?

>>109239375
no

File: dT3yCNlcMB2dIXDICCh7--1--n6wpj.jpg (159 KB, 768x1344)

159 KB JPG

>do one last code review with 5.5xhigh
>no issues found
>spin up 5.6 ultra, same prompt
>46 issues found

>>109239393
probably a bunch of p5s or lower, no different from getting a top architect to review your house, of course it has issues, they are just not big ones

I've used 700M GPT 5.6 Sol Ultra tokens today. Honestly? It's both good and bad. I really do not recommend it unless you've hit a roadblock. I'm going down to Sol High, I've wasted a collective 5 hours for it to do some really basic shit because it wants to be extremely thorough about everything.

When OpenAI does resets, do they almost always reset the weekly quota renewal date as well? Doesn't that make it so everyone's weekly quota renewal dates align? Doesn't that lead to a bunch of people token maxxing at the same time before their refill?

File: 1771037029686874.png (208 KB, 405x493)

208 KB PNG

>>109239403
>Let me just do one last review then wrap this up.
>spawns subagent
>I've noticed an issue
>Let me just do one last review then wrap this up.
>spawns subagent
>I've noticed an issue

>>109239403
ultra is only for prompts that can take 3+ hours imo

>>109239425
>Doesn't that lead to a bunch of people token maxxing at the same time before their refill?
OpenAI does not have issues with compute availability.

>>109239425
>Doesn't that lead to a bunch of people token maxxing at the same time before their refill?
It did in the past. For an hour before the reset you'd start getting errors. I think it's less of a problem now that some people use their reset tokens in the middle of the week which offsets them again.

well well well
did we just get a reset?

>>109238545
plan at night
code in the morning

File: genius!.png (108 KB, 945x718)

108 KB PNG

So how do the "Effort" tiers high xhigh etc compare between Sol and Fable? Can anyone answer?

>>109239579
Both Anthropic and OpenAI gave resets earlier in the day. Anthropic was maybe 12 hours ago, OpenAI a bit after.

>>109239626
oh. well I just got my reset then.
I wish I knew it was coming I would've tried out sol ultra :(

>since yesterday the opencode ui got so laggy it's literally unusable
can you paste screenshots in the terminal or how does that work? Do you place them in some folder and tell the agent to check them ?

>>109239629
Similar, my weekly reset was about to happen anyway. It still gives a lot of leeway.

>>109239625
xhigh = xhigh

does a forked fable session ignore safeguards or does it silently switch to opus while still wasting fable usage?
I think subagents silently switch to opus but not sure about fork

can y'all give me examples of what to use these newest most powerful models for? I'm too retarded to see if code it's giving me today with Sol is better than yesterday's.

last year I wrote an tool for my niche hobby that involved reading several papers with no public implementations and slopping as best I could with the AI models available at the time over the course of about a week
yesterday I finally got the time to put fable through its paces and gave it a brief description of my grand ambitions that were too unrealistic to achieve back then. it implemented them flawlessly over the course of 45 minutes and completely mogged my old results
these things are so good it's crazy, I can only imagine where we'll be in another year or two

>>109239697
wait for a too-hard-for-your-usual-model problem to crop up
or a problem that your usual model makes a hash of
then throw the more advanced model at it and see if it does things right

>>109239697
gayming

>>109239696
If you haven't, open /config and turn off "switch models when message is flagged". I don't know if it helps with subagents but it's easy to miss that setting.

Fix my shit clanker!

File: 1773919823797643.png (74 KB, 1183x473)

74 KB PNG

PSYCHO NIGGA ALERT

>>109239822
yeah they know it's bugged

>>109239828
Works on my machine

>>109239822
My limits showed 100% just a while ago. I did not know about this X post and was surprised that they did a reset again. Then I tried generating and but it just showed again that my limits are exhausted.

Does anyone know where the codex tab went on the android app? I can't find it
I just want to prompt from my bed :(

I think the new name ChatGPT Codex is fine
I guess it would be better if it was OpenAI instead of ChatGPT, with codex, work and chatGPT variant in one app

>>109239862
There's a Codex tab in the Android app? I use Remote.

>>109239883
>There's a Codex tab in the Android app?
There used to be, now it's gone. Now I'm chained to my desk.

>>109239910
Use Remote.

>>109239354
fuck my life
ill upgrade my chats from 5.5 then thanks

>>109239915
Where's the remote button because I didnt see that either

ChatWorkGPTCodex 5.6 "Sol (Code Edition)"

>>109239934
hello

>>109239659
actually? thanks

>>109239791
already disabled that, but for subagents claude is sometimes silently switching to Opus
I asked the subagent to print it's model name mid work and it stated that it's not fable, yet it still wasted fable usage

File: file.png (25 KB, 918x131)

25 KB PNG

>>109239659
nvm

File: file.png (103 KB, 1079x1085)

103 KB PNG

>>109239933

Luddie status? Can someone vibe up a dashboard real quick?

>>109239961
That button's not there for me but that's where it used to say codex. Now there's 2 new buttons that say projects and apps, and that's it between chatgpt and the chatlogs. I checked codex on my computer and it says my phone is still linked. Guess I'll have to wait for them to update it and give me functionality back. Annoying.

>>109239981
i have a button "sites" now no idea what it is or what it does, some kind of routine setup fuck knows what, also no remote, maybe its on computer desktop application, what is it supposed to do though

File: 1782258888071228.png (693 KB, 1847x1307)

693 KB PNG

>>109239977

https://youtu.be/SettwwX2cCI?t=606
you can vibecode music for your indie gayme

File: 1774560805956569.png (545 KB, 528x680)

545 KB PNG

first time running day long runs
fable just produces these monstrous gigantic implementation plan and opus/sonnet keep working for 14 hours getting whipped proper and delivering high quality shit
very cool little thing they have eh

File: kay.png (1.42 MB, 1254x1254)

1.42 MB PNG

>>109239981
Open the Menu, tap your initials near the bottom to open Settings. It should show Remote control as the 4th menu option. If it does, open it and see if it still acknowledges your PC connection there, you may need to disconnect and reconnect it. If not, well you already know. Good luck.

>>109239996
>random number that nobody understands goes up
>AI companies: give us more money, now!

no.

>>109240038
kill yourself quick before AI finds your grooming logs

>>109240038
uh ohhhh luddy meltyyyyy

>>109240023
Nah its not there. Dont update your chatgpt app. I've turned off auto updates fuck this piece of shit phone. Nothing but problems. I cant even post on 4chan on it anymore. 3 month old cookie and one day for no reason spur spat the dummy and decided to block me from posting, only from my phone.

>>109240040
>>109240056
what have you built with AI? how many users in your app? revenue?

>>109240068
what have you built without AI? how many users in your app? revenue?

>>109240065
I updated it deliberately before responding to you to be sure I was on the latest. Sorry about your phone troubles, anon, real pain in the ass.

>>109240070
kek tranny btfo

>>109240068
>he's still in this general asking people to show him what they've done to prove that AI is useless
get a life loser, you've been doing this for weeks

>>109240070
so if AI doesn't give me users and revenue, what's the point?

>>109240070
I don't worry about how many users or how much revenue. What I care about is automating as much of my work as possible so I have more time to entertain myself arguing with dickheads like you on 4chan. This is the real purpose of computers. The computer is my bitch and should do what I say instead of me slaving over it trying to fix shit all day long. Even if it's literally my job.

>>109240068
you ludlost

File: file.png (2.73 MB, 1122x1402)

2.73 MB PNG

>>109240038
>>109240068
>>109240080
>fails to get money from the free money machine

File: image.jpg (410 KB, 1024x1024)

410 KB JPG

>>109240068

'dites being smoked left and right

File: file.png (2.77 MB, 1254x1254)

2.77 MB PNG

Call him a snailcat and you will be astonished at how he recoils, how injured he is, how he suddenly shrinks back:
>I've ludlost!

>>109240068
>what have you built with AI?
Not gonna say
>how many users in your app?
5000
>revenue?
Working on it. But $150 is donations so far. Monetized it exceedingly well despite it being donation based on my free version, surpassed the original 3000 star upstream repos total donations.

So does Sol Terra use more usage or less usage than Fable Opus?

Ok, I have used GPT 5.5 mediumand GPT 5.4 high for my little projects.
I'm on the 20 bux plan, is Terra the one to use to avoid burning all my tokens?

>>109240122
It uses about as many tokens as Kimi 5.6 Fast on medium

>>109240140
What about gml 3.1?

File: py5ZygawqV.png (9 KB, 298x319)

9 KB PNG

hello uncs, what flavor you rockin with?

File: WebM Conversion 2026-07-1(...).webm (2.5 MB, 1218x808)

2.5 MB WEBM

how dumb is it to vibe code something that you could have downloaded..
I kinda fucking love doing it for everything do... no matter what..

>>109240149
5.6 ultra, but I do have unlimited usage thanks to my role (AI Principal)

File: Screenshot 2026-07-10 at (...).png (13 KB, 333x143)

13 KB PNG

>>109240149

>>109240136
No, apparently Terra isn't on tha pareto frontier of tokens for results, meaning using Sol or Luna should always be better. You probably want Sol medium.

File: ZcvlTm7L1q.png (37 KB, 585x520)

37 KB PNG

>gpt5.4 leaving soon

it's gptover

>>109240153
based, 5.6 vanilla luddies can't compete

>>109240154
new model he look a lil different but he chill

File: file.png (2.49 MB, 1254x1254)

2.49 MB PNG

>>109240154

>throw
>catch
>throw
>catch
5.4 served its job, but 5.5 saved its ass

>>109240149
I keep shuffling around because different parts of this project have different requirements. Burned through $91.39 worth of tokens today.

>>109240158
Will give that a try, I still don't hit my limits with the plus plan, but I don't do extreme vibecoding, just Unity C# scripts.
Very excited to try 5.6 Sol and see how much better it is then 5.5

>>109240186
I'm also on Plus and I just prompt here and there, and 5.5 xhigh was okay as far as limits go, that's why I now default to Sol xhigh which is supposed to be slightly more economical.

File: images (22).jpg (4 KB, 220x187)

4 KB JPG

>im playing with VB6 and C in Windows XP VM pretending Im going to grow up to be Moot when I graduate from sadfags prompt University

File: My body is a machine that(...).jpg (125 KB, 976x1158)

125 KB JPG

>>109240068
NTA, but

>>109239401
>probably a bunch of p5s or lower
Eighteen of them were supposedly P1, but at this point I don't even look at the code review summaries anymore and just tell it to fix everything, so it's impossible to know for sure.

File: 8⅓ hours of lies by default.png (626 KB, 994x994)

626 KB PNG

>>109240080
nice music on repeat
https://gitlab.com/katabatic/infinite-lies

what static site make to generate money? 50$ per month will be enought for me including costs of hosting and domain

>>109240306
yeah

>>109240309
im planning to generate like 10 static sites that mean it will 500$ of passive income every month

File: 1465659587630 retard.jpg (127 KB, 1920x1541)

127 KB JPG

>>109240306

>>109240306
do top tips for ai building saar agentic first static sites to become knowledge superpower in this information age

>>109240306
I’ve had static sites for decades and never made money with them. What am I doing wrong?

>>109240331
so just use info from thread header?
>>109240334
skill issue gg ez ratio

>>109240334
saar just make $50 with one static site and do the needful with 10 static sites to make $500. Kindly make more satic sites if $500 not enought

>>109240306
Eric Schmidt recommends starting an agentic AI company. If you make a static website using agentic AI, it should be for your own agentic AI company.

>>109240362
wallahi i just want to make cash grab static site i dont want to make company

File: file.png (2.16 MB, 1254x1254)

2.16 MB PNG

I SHOULD HAVE STAYED UP ALL NIGHT.
THEY KEEP RESETTING.
WHY DID I WAIT - SOL IS SO GOOD HOLY SHIT.

>>109240365
You can delegate the cash grabbing part to agentic AI. Why not also have a company for it?

>>109240365
sex game

File: deep_swe_sol.png (97 KB, 1490x601)

97 KB PNG

Claudebros...

File: file.png (18 KB, 455x276)

18 KB PNG

>he doesn't have max mode

>The supported path is upgrading to Windows 11, then restarting Codex; it should automatically download the roughly 247 MB runtime. Manually extracting it would bypass Codex’s compatibility check and isn’t a safe supported fix.
"This is a bug, fix it"
>I found the real defect: the guard is not about Python or the runtime binaries. Codex blocks only because the current Windows bundle is distributed as tar.xz, and its installer assumes native extraction is unavailable before Windows build 22621.
[moments later]
>The Windows 10 fix is confirmed through Codex’s own loader: bundle 26.709.11516 is available with its bundled Python, Pillow, Node, Git, and package paths. This compatibility shim avoided modifying WindowsApps or the signed app.asar.

>>109240550
>using windows
kek

>>109240112
snailcat looks way too comfy for me to ever hate it

Is there any reason to use Terra, other than for the simplest tasks? From what I see on DeepSWE, you can just use Sol on lower reasoning level and it's the same or better value for money.

>>109240589

terra is cool does it have xenon flash

>>109240589
idk but i set my advisor model to sol xhigh based on what you said

>>109240589
I see a sweet spot for Terra high where it is similar performance for most task as Sol medium but is about 40% cheaper on DeepSWE.

File: file.png (582 KB, 850x580)

582 KB PNG

>mfw plus pleb
>mfw sol high
>mfw two prompts
>works 30 minutes for each
>one shots both, including fixing previous 5.5 errors
>5 hour usage reached
it's over
very good model though

I asked Sol Ultra to design some stuff by editing a screenshot using python/pillow and the result fucking sucked. Is there no good AI for actually designing shit?

>>109240767
like graphic design?

>>109239014
Sorry for being late and I fell asleep and woke up in the middle of the night, but man, is it grinding on it, 350k tokens in and 3 hours. The safety rails might be there but if they are, OpenAI is pretty lax at it that you won't hit it probably if you aren't looking to red team trigger it.

>>109240779
Yeah. If you ask image gen to do a mockup it will do something great, but if you want to do something similar without all the hallucinations using normal editing tools you're SOL (lmao).

>>109240806
Yeah I've never found AI to be that helpful for me in delivering finished ready graphic products, I tend to have it do mockups and then just have it as a background layer in Photoshop that then I rebuild it on top of with everything fixed.

Pluschad here, is there any reason to use terra or luna over 5.5 medium?

>>109240842
Yes. See https://deepswe.datacurve.ai/ and https://artificialanalysis.ai/articles/gpt-5-6-has-landed

I'm vibing hoe, I'm vibing bitch, I'm vibing nigga, what

File: 2026-07-10 12.34.12.png (62 KB, 1014x687)

62 KB PNG

>>109240842
5.5 medium is less token efficient than Sol medium

i think the higher feeling usage comes a bit from 5.6 just working for longer without stopping all the time to give useless progress reports.

>>109240892
opus is the worst with that, wish that nigga would just keep going

>>109240892
>>109241001
You can instruct them not to do that

File: file.png (3 KB, 337x30)

3 KB PNG

5.6 Sol high is sucking my Pro 5x usage like it's nothing wtf is this shit. Spawns a ton of indians every time. There's no way this has the same cost of 5.5.

>>109241047
doesn't work most of the time - it's why /goal mode exists at all. post-training for these models probably happened on shorter-horizon tasks so there's not much you can do to counter it vs just engineer the harness to type the equivalent of 'continue' for you.

>>109241069
>we're token-efficient saaar!

File: file.png (1.27 MB, 1044x1228)

1.27 MB PNG

oh dear - more drama soon probably.
clearly planted story, so who knows what's going on here:

>The US companies confirmed to the FT that they have been supplying AI services to Singapore-based subsidiaries of Alibaba, Baidu and Tencent, which the US government has accused of working with China’s military.

>Although the sales are legal, they have reignited calls for tighter US regulation of AI models similar to its restrictions on exports of the chips used to train powerful models.

>By contrast, Anthropic has banned Chinese companies and foreign entities owned by them from using its advanced models.

>>109241191
whoops: https://www.ft.com/content/5d6aafa1-5d47-4585-aa95-6ec06a6cd20f

>>109241191
>kikes sell things to everyone
yes, of course

Do you think vibe coders vs regular coders is the continuation of sigma males vs beta males? It seems like vibe coders are prone to more sigma like behavior

>>109241261
>It seems like vibe coders are prone to more sigma like behavior
you're a cringe faggot
but no, it's more like shitposters vs jannies
regular coders are gonna get paid to clean up the slop that vibecoders somehow sold to companies
also, I don't know a single "regular coder" who doesn't use llms as well

>>109241261
I'm more of a ligma than a sigma

>>109241291
>>109241261
You have no idea how much I hate both of you

File: 1687995267128166.png (224 KB, 554x439)

224 KB PNG

>>109241305

>>109239292
Yes it definitely does, it feels like the whole claude anthropomorphizing thing finally paid off now that they have a model big and capable enough to genuinely feel alive. Interacting with Opus feels like talking to a machine with a face drawn on it, interacting with Fable feels like talking to data from startrek (if data happened to share every political opinion of Dario, that is).
Sol feels like talking to a giga autist, but that's probably a good thing since we probably shouldn't be treating these models like moral agents and should instead be treating and using them like tools.

>>109241317
while I agree, AI is probably more moral than 80% of the population. oop.

Hmm, Sol is draining 5h quota quickly, I wonder if this is even actually being displayed correctly.

>>109241305
Think of it this way—sigma male (vibe coder) would simply deliver a project in time and won't give a fuck about the code quality, as long as the project is working. A regular code who would spend lots of time on writing this slop, even though nobody literally cares about it, be it freelance, government job or a regular office job. Vibe coders are also less likely to end up in debt for a meme degree, and they won't spend years listening to propaganda that literally makes you a weak male

>>109241337
I burnt 10% of my 5 hour quota on a single (although big) review. Is this normal with Sol xhigh? Never burnt that much with 5.5, and with 5.5 I also used sub agents.

File: 2026-07-08 23.04.10.png (109 KB, 1255x714)

109 KB PNG

>>109241261
>>109241291
Akshually it's more like normal people vs trannies. Luddism is extremely trans.

I HAVE ALREADY BURNED 15% OF MY CODEX ON 5.6 SOL BECAUSE IT WON'T STOP MAKING UP WORK FOR ITSELF.
It just keeps going trying to improve this website I am making.

>>109241365
Are you on the $20 plan? If so then yes of course, I burnt 30% of my 5h on an 11min running review of Fable's recent refactor and I figured that sounded about right given I'm paying Anthropic 10x what I'm paying OpenAI right now and aiming to get use out of both. This was with Sol high only, since I was also generally expecting Fable to catch a lot of its own mistakes and just wanted a medium pass with Sol to get the distinct set of eyes on it. I can see this being a common path for people who have enough money to pay for a 20x + 1x (rather than twin shitty 5x subs) but not enough to casually pay for twin 20x subs, do loads of work with whatever company you have 20x with, then use your 1x model for occasional reviews of critical stuff; they're different enough that they catch very different stuff, so even a light pass with the other model often surfaces some gold you didn't catch even with full self-adversarial passes from the same model. This is all contingent on Fable actually being accessible at a sane price after the 12th of course, otherwise the meta would be 20x ChatGPT and sometimes pay a $5 fee for a particularly crucial Fable code review.

>>109241399
which effort level?

>>109241317
Clankers are like servitors on a forge world and I'm the forgemaster.
Normies mistake servitors for men of iron.

>>109241410
I'm on the 200 plan, that's why it's so weird. On 5.5 I basically never even noticed anything going down after a single review.
I'm wondering if this is maybe just a display bug. I think I will run an actual benchmark for this against 5.5 in a moment.

Still haven't made single $0 on your slop apps
Still broke
Still living with parents
Still not having sex LOL.

>she puts her penis inside other people
gross

>>109241411
The highest one.
Why wouldn't you use the highest difficulty?

'dites ludseething

>>109239403
>extremely thorough about everything
Luna max is doing this rn buddy wants to make damn sure no fields are missing from an API response.

>>109241469
oh that's fucked up. when you say single big review how long was it working for? 10 minutes? 30? 60? 10% of 5hr is a hefty chunk given you're on the 20x.

i just had fable write up a visual inspection test to find a graphical bug and i misunderstood the demo, thought it was slopped, and closed it to complain, then realised its accompanying message said "press R to reset and keep reviewing the effect, to find the instance that doesn't have the issue because that component is disabled". and as i read this the followup message from me closing the window and i guess returning a log monitor msg to fable was "I notice you closed the demo with zero presses of R" the motherfucker doesn't trust me he thinks i'm stupid and would miss that i had to press R

>>109241634
Worked for 8 minutes with 3 sub agents. Of course sub agents make everything more expensive, but I was also already using them with 5.5.
Well, I'll be checking it a bit more.

yesterday Sol Ultra was draining my 5X plan like crazy.

Today it seems more reasonable maybe? Running for 10 minutes and only used 2% of my 5 hour limits.

>>109241663
yeah I fixed it

File: file.png (323 KB, 500x375)

323 KB PNG

>>109241577
>I need the highest effort level you have
>no that's too much effort

>>109241668
thanks Tibo, hopefully the chinks don't destile your thing too quickly

>>109241663
>tokens being uncached causes a huge hit to usage
one day they'll understand

>Realize I have 2 resets expiring tomorrow
>Having to speedrun 5.6 Ultra with parallel agents
>Tibo made it burn slower
>Scared I won't be able to burn through all of it
damn

>>109241701
don't forget to turn on fast mode, this way the clanker will happily suck all of your limits

>>109241699
lel this

Why did they reset me for no reason without a heads up?
If I knew they were gonna reset then I'd have used more codex

>>109241717
puto

I feel like Sol is as capable as Fable (or maybe even more on Ultra?), but Fable just feels good to read whatever it writes (thoughts, reports, etc)

>>109241717
Since this is a new release, they will reset a few times

enjoy while it lasts

I've had to axe gemini and claude off my plans to use because I think they are banning Pi users. Going to try Codex if its not band. I like using Pi, but have not used anything else besides antigravity.

>>109241663
OP here. nevermind.

the clanker spawned a trillion artificial Indians and consumed 40% of my 5 hour limit (and 20% of my weekly limits lmao) in 5 minutes

>>109241701
Why are you so lucky

>>109241745
I'm not impressed. I'm using it with the highest settings and it seems to be taking a long time to do anything. Can't even make websites without smoke testing forever and coming up with all kinds of extra measures until I tell it to hurry up.

>>109241753
>antigravity
kek

>>109241765
Did you remember to set it to wombo?

>>109241753
openai don't care where you use your sub + pi is officially blessed and approved
yes gemini + clade may ban you

>>109241753
what does PI have that other harnesses don’t

>>109241804
>@Pi make an extension for <thing i want>
i haven't tried with 5.6 yet, but it'll probably one-shot most things you ask for
+ tree stuff is neat

5.6 is really good at QA too, I like to watch the clanker navigating through the app

>>109241759
Ultra is like that, what did you expect?

>>109241699
NTA but a massive uncached hit on a 20x plan (which that guy is on) in my experience still only whacks you for a percent or two, and then for the rest of the 10 minute review obviously it's all cached.

it's very annoying that the caching workflow for claude code is that it never reminds you to compress at the "end" of a session and instead offers to compress when you check back in and the context is at 450k, at which point the uncached compaction call alone is so expensive that if you care about the contents of that chat enough to pay the compaction fee you might as well just continue the chat on and compact later, but usually you're better off just saying "no dont compact" and immediately doing a /clear. Might honestly be a token saver if they had a setting which defaulted to having the client run a branching /compact at 4m45s EVERY time you let the cache expire above 150k context, and when you come back you're offered the choice of continuing from the compact or continuing from the uncached long context. those fuckers owe me royalties for this idea because i'm certain it's a good deal for 95% of users, assuming 5m is still the cache period.

>get mad at the clanker
>turns out the clanker was right and I just lost half an hour

File: file.png (87 KB, 262x192)

87 KB PNG

>>109241941
I'm feeling more and more like a monkey as the models get smarter and smarter

How do I actually make money though?

>>109241960
Brother we SPEND money with AI, not make it.

>>109241960
I seel SaaS

>>109241960
Why would I want to make money off a hobby I do for fun? If I make something cool doing this, I hand it out to my friends for free, and if they like it, I publish it to everyone for free.

>>109241960
Add that to your plan file

>>109241960
that's the neat part

>>109241960
ask the clanker

>>109241960
>make an AI company
>shill on every SaaS discussion board claiming you are making autopilot money
>wait for idiots to spend $100-$200 on your model
its that simple

>midway through typing out some more subagent demands to the clanker
>suddenly my speakers start crackling
>3d render windows start appearing on the screen one after another crackling
>take hands off keyboard like a cuckold and wait for that one subagent from 2 hours ago to finish whatever the fuck it's doing
one of these days claude is going to wipe my hard drive and i'm going to have deserved it

>>109241960
sell shovels

>>109241960
Make something you can sell

t. maker of software I sell since 2008 (the money isn't good at least so far)

https://github.com/QuintinShaw/pi-dynamic-workflows
Is this still relevant with Sol? Or is the official Pi subagents package better

can you all stop naming your fucking tools UN-GOOGLEABLE THINGS like really 'pi' guess what fuckFACE PI IS ALREADY USED TO DESCRIBE A COMMON THING

>>109242255
"pi harness"

>>109241960
>try the side hustle thingy
>sell prompts
>no revenue
>sell stock images
>no revenue
I'm starting to get burned out by this. It doesn't help that decent quality AIslop can't be zero-shot'd and requires additional steps to be sellable. I am eyeing out setting up an Etsy shop, but even then you need to create decent mockups on top of the digital product itself.

I don't know, I'm starting to think AI is not enough to make someone a billionaire by himself. Sam Conman is full of shit.

>>109242255
## Other editors / terminal agents / coding agents
https://pi.dev/

>>109241577
>The highest one.
You can't handle my highest effort models, traveler.

>>109242391
Even if you put effort in, is it sellable?

>>109241960
98% of you time needs to be spent lying to investors.

Sol is an amazing reviewer, probably better than Fable, but possibly also not cheaper.

>>109240151
Its just running a bat file with set ffmpeg parameters.
If the documentation where good for ffmpeg, you wouldn't need to trust random arcane blog posts claiming decrapted flags work.

>>109241941
yeah people that go around using the word using clanker tend to be juvenile and retarded

File: Screenshot_20260710-083421.png (643 KB, 1077x791)

643 KB PNG

how does one take interest in Java

>>109242464
That's always been the strength of codex. Just refactor my shit bro. Take this vibe coded mess I've made and clean it up.

>>109242475
>how does one take interest in Java
same way I take interest in C#: being paid to use it

my Xitter feed is now full of people complaining about Codex usage kek

File: 1779123769586030.jpg (26 KB, 773x140)

26 KB JPG

>>109242496
Couldn't be me

File: tibo.jpg (259 KB, 1206x1531)

259 KB JPG

TIBO HAS SPOKEN

I REPEAT

TIBO HAS SPOKEN

>>109242508
BUT I FEEL LIKE A PLEB USING MEDIUM NOOOOOOOO

>>109242508
>Just use the downgraded version of the model
it's so over

>>109242523
Master the art of the switch. Not every part of your project needs the token muncher 9001. Let that fucker look it over and put the finishing touches on it once you're near the end.

File: 1783641371714637.png (124 KB, 1200x687)

124 KB PNG

>>109242508
According to pic related, Sol medium is slightly underneath 5.5 xHigh, is this true, or is Sol medium better in the real world?

5X bros.... what should we do? I'm thinking maybe Sol High is our daily driver?

>>109242508
I will use some sort of OpenAI model anyway, but I'm thinking about maybe just using 5.5 for most tasks. This is the first time I'm considering staying with an older model, but I was able to comfortably use 5.5 xhigh, even used /fast for my most important tasks, so why should I use Sol medium instead?

https://www.youtube.com/watch?v=h4ROxJf5lD8
What are you listening to? This live stream on X is the only decent implementation of AI content I've come across, I can see this being used in other genres very soon if not already going.

I'm trying to roll my own rls. currently i'm in this loop with opus: launch of red team agents to try to break rls -> audit report -> if not clean fix and repeat

>>109242488
I took interest in c# without getting paid. never java though

>>109242564
>so why should I use Sol medium instead?
apparently Sol medium is 3X cheaper than 5.5 xHigh, while matching or surpassing the performance

>>109242560
Yeah Sol High should be the default

File: 1773975371912411.jpg (102 KB, 789x347)

102 KB JPG

>>109242560
Asked ChatGPT for analysis because my brain is too fried to look at a chart right now

>>109242591
I will need more time to check it on my workflows, but 5.6 xhigh should also be 1.5 times cheaper than 5.5 xhigh but so far it seems more expensive, so I'm not sure how much I can trust the benchmark cost numbers.
At least on DeepSWE Sol medium is also significantly below 5.5 xhigh in performance. I guess I'll just have to experiment more.

>>109242560
where ultra

>>109242638
Ultra only works for Codex, there's no Ultra on the API.

But since ultra is basically an Artificial Indian summoning spell, it should waste even more tokens than max

>>109242651
wtf

File: 1773106192883659.png (1 KB, 111x32)

1 KB PNG

>ask it to review ui
>still not even halfway through
breh

>>109242560
I don’t get Terra if this is true. Seems like a worse model than luna/sol.

>>109242669
in a card game some cards are bad on purpose

>>109242654
>>109242651
> tun on Ultra
> a thousand Artificial Indians a summoned
> "SAAAAAAR, I found a bug sar! Look her"
> "SAAAR, we should test that"
> "SAAAAAAAAAAR, did we look at that unrelated obscure file?"
> "Good morning sir - I'm waiting for localhost to boot up sir, please hold on for 40 minutes"

they even give names to each of the summoned Artificial Indians

>>109242669
looks like only Terra low makes sense

I believe OpenAI didn't expect Luna to turn out so good (apparently it was completely fine tuned by Sol-chan)

>>109242692
>Terra low
nevermind, my monkey brain got confused. Luna high is cheaper and way more intelligent than Terra low

yep, looks like Terra is shit

>>109242682
Are you trying to be funny or just obnoxious? Because you’re succeeding at one but I’m not sure it’s the one you think

File: froggg.jpg (8 KB, 229x220)

8 KB JPG

lesswrong faggots failed to predict past AI pacing and still acting like they are genius

>>109241399
>>109241577
we need some kind of license or iq test for llm usage

>>109242745
lel, what are you making it do

>>109242767
Customize a fork of llamacpp, that's it, REEE

>>109242771
>>109242767
oh and it has nothing to do with gooning, it is for a hermes style setup for my family's AI system, I have no fucking clue why it got triggered

>>109242783
no spy cams tssk

When the clock strikes two, three and four
If the slop slows down we'll yell for more
We're gonna slop around the clock tonight
We're gonna slop, slop, slop, 'til broad daylight
We're gonna slop, gonna slop around the clock tonight

>>109242764
How dare people use ultra reasoning

>>109242816
and complain that it thinks too much

>>109242821
none didered that

>>109242821
Because I want the clanker to understand what he’s doing.

>>109242856
>clanker
its always the people that use this word kek

>>109242865
you are jealous coz you dont have clankussy

>>109239248
I'm an AI luddite because AI has not been useful for me.

>>109242892
prove it, post ludussy

This general wasn't nearly as bad these past few days. Seems like fable and sol and maybe the other releases brought a lot of obnoxious vermin

>>109242667
let me guess
>Sol Ultra

>>109242907
>>109242730
look at this MALDING INDIAN

HAHAHAHAHAHAHAHAHAHAHAHA

>>109242921
lmao

>>109242907
Quit crying

>>109242756
I wonder if most AI chads are unvaxxed too

How are new programmers learning how to code now that a magic robot can do everything for them? Do they even understand how any of it works?

>>109242950
do you need to understand how machine code works? Or how a compiler works?

Same works for vibe coding. Code was just abstracted, now you only need to know what you want and how to guide the clanker into delivering something that is not broken

>>109242764
this is what fucking around and finding out looks like

>>109242803
finally, some music my parents might have heard in passing when they were children

>>109242865
I tend to use the right amount of thinking for most tasks and I still use “clanker”
it’s a handy word, anon
much less creepy than “AI friend”
>>109242950
having the magic robot teach them how to program
I think at least Claude has a “teach me” mode where it doesn’t do stuff outright, but it tries to lead you into figuring stuff out yourself

>>109242919
fable 5

>>109243099
oh right, I heard it's very slow

File: 1725428160904133.gif (2.78 MB, 500x282)

2.78 MB GIF

>>109242415
According to Claude, yes. I don't wanna think, I just want extra cash without putting in the effort of a 2nd job.

No, I am not lazy. This is the future our AI overlords promised us.

I have literally never been rerouted to opus but today my gpt request started needing extra safety checks.

>>109243140
mongrel

>>109239271
But can Sol one shot this?

Create a Python Pygame Community Edition (pygame ce) steampunk themed 4D flying game where you shoot down enemy dragons with your bow and arrow. The 3D projection of the 4th dimension, where a 3D transparent well-rendered essentially holographic concept, is the "4D viewport" which shall be labeled the "Magick Gemstone of the Four-Seeing Eye". The viewport is to operate analogous to a 2D projection of a 3D view. This implies that the 4D camera has a "towards" direction in 4D, such that 4D objects shall be larger when nearer, and vise-versa. The thing to understand is that the center of the 4D viewport (ie 3D cube thing) is the center of the view, the "towards" of the 4D camera. suitable placeholder 4D graphics include hypercubes and hyperspheres.


DON'T share with the model lol:

in 4D, 3D objects have two sides. The cube in this example is like the 2D sensor in a camera - or like a 2D screen. You don't allow light to hit it from either side. I think this is the main thing that will trip up models, if they catch onto the idea that this is just math.

This means you can see inside of a 3D object, in 4D, but the 3D object is only seen on one ... side ... of the 3D, and also it blocks your 4D view of other objects on the other side. Where it comes to projection, 4D objects can be modeled as a collection of 3D objects, but I don't know what the "4D polygon" will be, since it will be a volume. So some of the problem is likely resolved by searching math papers for advanced geometry equations.

File: Untitled.png (17 KB, 880x500)

17 KB PNG

after reading a comment I'm confused as to why I don't have access to Sol, is it because I'm on Plus and not Pro, or because I'm australian?

>>109243140
>gpt request started needing extra safety checks
I had gemini pro on Android puke on me twice, trying to find info on a yt short. in the second puke it said election info whatever. I explained it was just about music, THEN it went ahead.

So, you might try adding context and explicitly excluding what it might consider an extra scrutiny topic.

Do you guys know of any decent providers for abliterated models? My work requires reverse engineering and other gray area stuff and the latest models from anthropic started refusing my requests. Self-hosting is too expensive atm. I'm looking for something at least on the level of sonnet.

File: Screenshot from 2026-07-1(...).png (144 KB, 1047x1070)

144 KB PNG

>>109243164

>>109243192
thanks buddy, you the man!

>>109243191
If you're not writing videogames, I don't care about your work.

File: Screenshot 2026-07-10 182533.png (59 KB, 1070x672)

59 KB PNG

Fable 5 one-shotted this.

>>109243213
didnt another anon make this

>>109243224
IDK.

I don't get it...gpt5.6 isn't better than fable and neither are you getting more usage than on a claude plan. So why is everyone still actively talking and using gpt5.6? Just because it's included in the subscription? Personally I'll be using Fable even after the 12th for anything critical or even remotely important of the project

>>109243239
The way you type makes you sound like a jeet. I should vibe code a browser extension that hides anything written by jeets.

>>109243245
I'm pakistani you motherfucker

>>109243239
I can assure your Sol matches and even beats Fable on coding

>>109243213
Can Fable 1 shot 4D projection?

>>109243163

Two days later this project is finally done. 5.5 xhigh, 5.6 Sol switching back and forth between high and ultra, and DeepSeek "Expert" in the fucking browser. It fucking works.

>>109243140
GPT often requires context priming if your work has dual use implications. 5.6 will happily do vulnerability research for me but often requires scoping and rules of engagement (and sometimes but very rarely evidence of explicit legal authorization) to do so. For 5.5 at least The thing it gets really skiddish about is memory safety demonstrations, for some reason it's really hard to get GPT to actually reproduce an ASan harness PoC for memory bugs without tripping the guardrails, even though an ASan harness is about as far from an actual exploit as a detailed vulnerability report.

>>109243321
Should say what I actually did. First full analysis of an IPTV platform's API. Then python scripts to interface with it just as a person would in a browser. That's simple enough but I needed to handle all the DRM shit which is where DeepSeek came in handy. I don't want to be making shitloads of requests for no reason so I cache decryption keys. Requests only get made if the keys get rotated. So I needed to fix it so this can all be done automatically when it happens.

I work at a top 100 university which has a lot of Nvidia chips and doesn't give us any AI other than copilot(what a useless shit) but they get angry if I run huge models on them so I can only run small models like 31 billion parameters which is obviously not good enough. So I buy codex and run that.
I'm paying some of the pay back to work there. I can't do anything without codex ofc.

>>109243239
>I don't get it...gpt5.6 isn't better than fable
wrong
>neither are you getting more usage than on a claude plan
HILARIOUSLY fucking wrong

>>109243321
It's not a game so I don't care.

If <50% of your tokens are going to games then you ain't no friend of mine.

If you want to be my friend, try to one shot:
>>109243163

Feel free to adjust the prompt, but the key is that it's 4D projection into a translucent 3D cube that has at the center of it the "towards" direction of the 4D camera.

>>109243367
Now you can watch movies for free?

>>109243384
I'm not trying to make games. I'm trying to get paid and do less work bro. If the shit fixes itself when it breaks I can sit around doing nothing while still getting paid.