[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: Buffcat by the pool.png (2.67 MB, 1254x1254)
2.67 MB PNG
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

## News
- (2026-07-09) I just made this thread last night before bed, jeez
- (2026-07-09) OpenAI Sol/Terra/Luna released
- (2026-07-08) Grok 4.5 out

## What “vibe coding” is, and how to do it
https://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 so
https://developers.openai.com/codex/cli
https://claude.com/product/claude-code

## Not worth it for code, but maybe good for other things
https://geminicli.com/docs/
https://x.ai/cli

## Open / local / self-hosted
https://github.com/OpenHands/OpenHands
https://github.com/QwenLM/qwen-code
https://github.com/QwenLM/Qwen3-Coder
https://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF

----

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grilling is a favorite

## Other editors / terminal agents / coding agents
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://ui-ux-pro-max-skill.nextlevelbuilder.io/
https://stitch.withgoogle.com/

## In-browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://docs.github.com/en/copilot/tutorials/spark
https://v0.app/docs

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109232574
>>
damn, fable is incredibly slow now
>>
>>109236005
makes sense, Fable chugs tokens compared to Opus
consider having Sol figure out what to do and have Terra models actually do the work
>>109236010
so have it do a bunch of stuff and then go for a swim
>>
Hey, wait, whoa, I wasn't watching - it's on OpenRouter right now? I can use the API? Is it in the app/CLI or whatever? Is it finally time to resub? Are we going home?
>>
>>109236005
> Sol Ultra chugs tokens like there's no tomorrow, Jesus. Many times what 5.5 xhigh did.
lmao, this MUST be a bug

> t. have 5X Pro plan
> turned on plan mode (Sol Ultra)
> 12 minutes thinking
> 5 hour limit went from 10% to 87%
you can't make this shit up
>>
>>109236047
should i be using out my weekly rn im at 89%
think theyre gonna do their reset?
>>
>>109236047
judging from my experience with Fable, that only MAY be a bug, especially if Sol is faster than Fable
>>
File: 1772975896835071.jpg (552 KB, 1708x1858)
552 KB JPG
>>109235833
>Quote from OpenAI on the livestream.
>'Already, Sol has been transforming our research program. As one example, GPT-5.6-Sol autonomously post-trained GPT-5.6-Luna.'
>>>>>its real
RSI SOOOOON
>>
>>109236068
isn’t that just distillation?
>>
File: file.png (28 KB, 905x185)
28 KB PNG
where is sol...
>>
File: 1761856275075562.png (235 KB, 1179x2556)
235 KB PNG
I'm gonna crash out HURRYUY UUUUPPP
>>
I was so hyped for 5.6 but now I'm just using Fable. Well played, Dario.
>>
>>109236089
jerk off, clean your room, prep some meals, then come back and it'll be there waiting for you
>>
>>109236081
no, distillation is training on output token
sol controlled the computer to train luna
>>
File: safety.png (15 KB, 1637x125)
15 KB PNG
>>109236047
found the answer! :)
>>
copex users are luddites at this point
>>
>>109236128
It gets the job done for cheap after payments to some guy in China. What more could I ask for?
>>
Anyone else finding Sol on Ultra super slow?
>>
File: slow-model.png (12 KB, 1014x67)
12 KB PNG
>>109236148
apparently
>>
> 5 hour usage limit slowly going down instead of going up

the fuck is Tibo doing? Just press the fucking button
>>
>>109236148
i am not finding sol at all
>>
120B A5B
>>
>>109236167
probably waiting for a full rollout so expect a reset in 24 hours
>>
sol flopped
>>
>>109236167
You need a Tibo?
https://www.youtube.com/watch?v=bWcASV2sey0
>>
File: file.png (14 KB, 644x169)
14 KB PNG
>Sol not included for pro subscribers
Whatever. I didn't even want it anyway.
>>
>>109236181
https://www.youtube.com/watch?v=A_HjMIjzyMU

F bonnie
>>
>>109236190
who is it available for...
>>
>>109236190
have you tried upgrading codex?
>>
>>109236196
first thing it asked me to do
>>
>>109236190
is Pro the $20/month one?
that makes sense — Sol will probably eat up your entire 5h allotment in ten seconds
>>
>>109236217
plus is the $20
pro is the 200
>>
>>109236217
not him but no the 20 one is plus. pro is 100 and 200
>>
>>109236095
The reset was expected but I will use them both. When I do I finally get 5.6.
>>
>>109236217
That is stupid.
Sol is apparently not better than Fable 5 and I can get about ~300k tokens from Fable 5/high.

>>109236229
>>109236232
My mistake. I'm on a plus subscription.
>>
File: 1763702191844952.png (2.4 MB, 1536x1024)
2.4 MB PNG
>>
>>109236223
not how it works.
5.5 at least was capped on the server endpoint
only 5.4 could be uncapped
>>
>>109236113
>>109236166
worse than fable
>>
>>109236217
Even fable is in claude's $20 plan. It's not much but still you can do at least some amount with it. Not sure why Sol wouldn't be in Plus?
>>
File: tpb82.jpg (421 KB, 784x1168)
421 KB JPG
IS IT IN THE APP YET? IS IT IN CODEX YET? DO I RESUB NOW OR NOT? SOMEONE TELL ME. I NEED IT TO SLICE A 3D PRINT.
>>
>>109236262
yes
>>
>>109236260
it'll be rolled out in 24 hours to plus accounts
>>
how is token usage vs 5.5?
>>
>>109236280
Fuck, man. Fuck.

Kate lives.
>>
>sol is slower
>more safetycucked
>not even included on most plans
Anthropic wonned.
>>
>>109236293
we already talked about that
it's terrible
but like the other guy I'm barely being able to use it anyway
>>
File: file.png (68 KB, 240x148)
68 KB PNG
>>109236298
>plans
>fable
>>
> can't find the context size UI in the new Codex app
They're still giving us 256K aren't they?
>>
An hour ago, claude was dropping shit for ten minutes straight.
I don't understand why Anthropic is shaking so much.
>>
>>109236316
353k
>>
>>109236293
impossible to know, it's unironically negative now (my 5-hour limit went from 87% to 45% while Sol Ultra is running for almost 30 minutes)

What I can tell you right now: Ultra is slow af compared to Fable Ultracode
>>
File: 1772392032695343.png (490 KB, 605x1024)
490 KB PNG
>>109236340
>negative now (my 5-hour limit went from 87% to 45%
>>
>>109236336
source?
>>
>>109236347
remaining as opposed to used
>>
File: 1757960986404257.jpg (45 KB, 452x442)
45 KB JPG
>invents some matrix math module I don't even need to be slightly more accurate
>routes everything though it causing performance issues and bugs. What it helps with will never even happen anyway.
>down to 55% 5h already
thanks Sol Ultra, very cool.
>>
File: file.png (38 KB, 531x641)
38 KB PNG
yay they did the thing
>>
Why did Google focus on speed for Gemini 3.5? No one gives a shit about them now. Do they even have a chance anymore?
>>
>>109236349
>>109235745
>>
>>109236362
>invents some matrix math module I don't even need to be slightly more accurate
top kek
classic GPT, I got that a few times with older models
I'm really cautious about letting codex cook on its own, I would be highly suspicious of ultra
>>
File: 1773254809888221.jpg (77 KB, 681x660)
77 KB JPG
>5.6 turned out to be a wet fart with 1/3rd the context of fable
>only a few more days of fable until i'm stuck with sol
ALL YOU HAD TO DO WAS MAKE THE CONTEXT WINDOW 1M AND IT WOULD'VE BEEN FINE WHY ARE YOU DOING THIS TO ME
>>
File: file.png (5 KB, 478x99)
5 KB PNG
>>109236366
I CAN NOT REDEEM SAAR
>>
>>109236375
keeek
>>
>>109236375
context is 1m on cli
>>
How does Luna and Terra compare to Sonnet 5?
>>
I just tried it and it's awful. Everyone switch to anthropic NOW
>>
>>109236403
do you realize your usage is going up by 50% soon, or did you forget after slobbering all over D'darrios dick?
>>
>>109236367
They only released 3.5 Flash, which competes with Luna/Sonnet. Everyone is still waiting for 3.5 Pro which is the big boy
>>
>>109236385
>source: my ass
>>
>>109236414
Buy an ad shill
>>
>>109236403
Ultra feels bad man, it's like spawning billions of Indians and they never finish your work
>>
>>109236385
I think only on API.
>>
>>109236366
>expires
wait what
>>
Test
>>
>>109236414
>your usage is going up by 50% soon
you're about as good at math as an llm
>>
>>109236425
get trolled
>>
>>109236425
You didn't know? We've been talking about this since they started. I frequently posted about how to check the expiration dates before it now being visible in the app.
>>
>>109236429
take as much time to cope as you need
>>
it's already pissing me off
>>
File: file.png (17 KB, 599x288)
17 KB PNG
>>
>>109236440
>calling out the retard for being retarded is coping
did you grow up on twitch or something?
>>
>>109236449
Man, some people are awful at this eh?
>>
why do I only have terra and luna still where's my sol I PAYPIG SO HARD SAM YOU FAGGOT
>>
>no sol yet
>chat stuck on 5.5
i'm such a plebeian
>>
File: file.png (14 KB, 613x108)
14 KB PNG
>>109236465
>>
File: 1783263667299086.jpg (207 KB, 1080x1428)
207 KB JPG
I think I'm not resubscribing. I think the well-poisoning Anthropic did to me is so complete that I'm just done with the Big Two Jews forever at this point, no matter what shiny new models they puke out. I hope 5.6 is good, guys. I'll be in China.
>>
>>109236449
>average CLI user
>>
>>109236415
No fucking way it competes with Sonnet 5.
>>
>>109236472
am 20x still no sol
i blame the gubbmnt
>>
>>109236477
qrd?
>>
>>109236475
see you tomorrow when you realize how good you had it
>>
>>109236482
on the app? I had to kill the process and restart it for it to download the update
>>
>>109236503
I'm actually selling my computer
>>
>>109236506
Got the update, cli and app I've got terra and luna but still no sol despite being 20x. I'll worry about it if it doesn't appear by dinner.
>>
>>109236503
I haven't been subbed to GPT/Codex for a while now. We had a falling out right around the time Dario announced he's going to be silently and invisibly sabotaging your code if it competes with Anthropic products. OpenAI does it too, they're simply not dumb/jewish enough to publicly announce they're doing it. Both companies make me feel fucking sick, like nausea in my stomach.
>>
File: 1770772835112403.png (2 KB, 207x61)
2 KB PNG
Alright just like Tibo said, I'm giving it my hardest problem. If it doesn't one shot this with le ultra meme thinking I'm denouncing the model.
>>
WHEN CODEX DESKTOP LINUX
>>
>>109236569
>I want to wash my car at the car wash...
>>
>>109236569
I too asked it to get me a gf
>>
>>109236569
why don't you write what you've given it and the result? even a meme shitpost prompt like >>109236578
would be better than this cryptic shit
>>
>>109236578
I want to count my car wheels at the strawberry washer
>>
>>109236223
If you don't isolate the Codex CLI app at a network level when it launches, ~/.codex/models_cache.json is always fetched from the server with the hardcoded 372k max_context_window.

>>109236316
>>109236349
Probably has same cap since it bundles the Codex CLI binary within.

7z x ChatGPT.dmg >/dev/null 2>&1

strings "ChatGPT Installer/ChatGPT.app/Contents/Resources/codex" 2>/dev/null | grep -oP 'codex-cli|OpenAI Codex v\S*' | head -2
codex-cli
OpenAI Codex v

strings -n 8 "ChatGPT Installer/ChatGPT.app/Contents/Resources/codex" 2>/dev/null | grep -oP '0\.144\.0-alpha' | head -1
0.144.0-alpha

strings ~/.local/bin/codex 2>/dev/null | grep -oP 'codex-cli|OpenAI Codex v\S*' | head -2
codex-cli
OpenAI Codex v

strings -n 8 ~/.local/bin/codex 2>/dev/null | grep -oP '0\.144\.0' | head -1
0.144.0
>>
>>109236584
It one shottedereded it, I was wrong to doubt Sam.
>>
>>109236584
>what you've given it and the result
Because I'm actually requesting it to make stuff for my company.
>>
my fable usage just got reset early... did yours?
>>
>>109236584
This thread is full of retarded jeets that think they are close to becoming billionaires
>>
>>109236596
fair enough
don't get why you would post in the first place tho
>>
>>109236597
...oh maybe it's GMT? very confused I know my reset day is Friday not Thursday.
>>
>>109236599
fuck you gora, post your prompt
>>
>>109236597
a reset dropped
>>
>>109236590
Is it really just a CLI restriction? I could even for the app if it's just that, I thought it's capped fully server side?
>>
File: 1762632076879689.png (52 KB, 360x360)
52 KB PNG
>subagents no longer have human names
>>
>>109236627
*fork the app
>>
>>109236607
nice thanks
>>
Is ultra the new xhigh?
>>
>>109236590
just use pi. when 5.5 launched pi default was 400k and they dialled it back.
>>
the nice thing is it looks like sol will happily work for hours on a request
it kept going even through a compaction
not sure if i speak to it regularly i can keep it going perpetually?
>>
oh no... immediately after I sent it a message it stopped
so i guess better left untouched if you run out of credits
>>
Did they just reduce the 5 hour limit by 90% or something why is one prompt making me hit the limit
>>
>mfw when i wait for the dust to settle instead of just refreshing till i have sol
>>
>>109236723
enjoy your stale quantized model
>>
>>109236723
I already vibed my million dollar app tonight, get left behind snailcat.
>>
File: 1782601319592054.jpg (78 KB, 1210x538)
78 KB JPG
>>
>>109236762
>fear
just plain old competition
enjoy it while it lasts, 'cause all these fuckers are vying for the monopoly/duopoly
>>
Has it been confirmed if the ChatGPT and ChatGPT Codex usage are shared now?
>>
>>109236762
Dario fucked up bad and he knows it
>>
>>109236762
yeah i don't think that's what that is
>>
>>109236824
What is it then?
>>
>>109236838
an opportunity for a reminder of how good fable is so you're disappointed when you have to use sol probably
i really hope i'm wrong and they allow us to uncap the context window to 1m because hoo boy this is gonna fucking suck if they don't
>>
>You’re out of Codex and Work usage
but my clanker doesn't afraid of anything he is still working
>>
>Sol still working after 41 minutes + Spawning 3 sub agents that worked for 12 minutes each
>Down to 47% Pro Usage (5 hour)
I swear to god if this doesn't solve the issue I'm gonna crash out. I would be angrier but I have 5 resets banked across multiple accounts.
>>
>>109236627
I tried. Even if you fork Codex and change its defaults, the OpenAI API still appears to overwrite settings for the model on the fly. You can see it happening in real time under ~/.codex/sessions, grep the latest .jsonl file for "model_context_window" and it's going to be 353400.
It's probably worth doing MITM between Codex and the API if you want to see more.
>>
>>109236780
lol they are not
>>
does sol have less context than 5.5?
>>
>>109236856
yeah codex is great for that, they let it finish tasks for ages, ive seen it spill over for 10+ mins past my usage running out. claude gets cut off on the next turn, very quickly, must be very annoying if you do big tasks and have the 1x plan.
>>
>>109236862
Ultra stands for Ultraslow
>>
File: file.png (20 KB, 742x141)
20 KB PNG
>i didn't run this on fable because i thought it would reroute me to opus
>>
Praise Sol
>>
My opinion on Sol Ultra:
> task: add a button and a new API request
> step 1: plan mode
> it spawned a billion Indians
> 20 minutes to finish the plan
> 40 minutes to execute the plan
so 1 hour to do this (I feel like 5.5 xHigh would make this in 20 minutes);

However, the clanker actually one shotted it, while 5.5 would always leave a small bug behind in this project (this app is a fucking bloated mess, caused by years of ancient luddity code)

Can't say anything about the limit usage because it's broken

so yeah, the thing is smart
>>
>>109236873
350K, so 100K more than 5.5
>>
File: file.png (2.08 MB, 1254x1254)
2.08 MB PNG
Heap salt at the sill. Let no warmth shine under the door. They come when the ground is wet. They are soft things, slow things.
>>
1min claude ad just played on fox during footy world cup hydration break
>>
>>109236991
did you make it?
>>
>>109237008
Huh? No. It started as an anti-AI ad. pretty creative
>>
>>109236262
>CODEX
Codex is ded. It's now ChatGPT App.
>>
>>109237040
Shame. I liked the codex name a lot more
>>
I just got into an argument with Claude. What a faggot. Even if Codex is worse at least it doesn't bust my balls with refusal loops nearly as much.
>>
> have $20 Anthropic plan
> have $100 OpenAI plan
> Sol Ultra drained all of my 5 hour limit than Fable 5 on Ultracode
Scam Altman is literally robbing us at this point
>>
>>109237055
was it because of cucked limitations on it's part, you was it trying to tell you you're retarded for even asking something like that?
>>
>>109237078
Cucked limitations. Any AI can get into what I would call the "refusal loop" where a single red flag gets raised and it becomes obsessed with it for the rest of the conversation and the only way is to start over. If you try to discuss nuance it thinks you're trying to skirt around the rules and continues being accusatory and judgmental. Claude is the absolute worst one for this.
>>
File: 1757450257470441.jpg (22 KB, 476x290)
22 KB JPG
I think this is an awful product decision
>>
>>109237099
it's confirmed: developers don't get work done
>>
>>109237099
lol
this is definitely a trap. I'm just not sure how it's set up
>>
> Sol-chan raped my 5-hour limit
it's over Copex sisters...
>>
>>109236895
He's still going
>>
Fucking FUCKS
FUCK YOU
Just fuck it
Just plug me back into the Matrix boys
I want the juicy steak
Just fuck it all
>>
File: 1771406186557892.jpg (36 KB, 386x518)
36 KB JPG
Sol Ultra takes a long ass time to do stuff
>>
>>109237178
>$20
How will you afford your McDonald's Fries subscription?
>>
File: file.png (25 KB, 2505x183)
25 KB PNG
why is my codex textbox grey now? It wasn't like this before. I want it back to black I don't like it
It was grey on my linux server, but it's always been black on windows, it's how i know which on im on. And blacck just looks better
>>
>>109237178
This makes me fucking sick to look at. I hate this.
>>109237193
Nigga I pay like $400/mo in API calls because I DO NOT WANT TO SUBSCRIBE TO THESE SLIMY JEW COMPANIES. I am richer than you, Gopal from Mumbai. Shut your brown mango-sucking lips, faggot.
>>
File: 2026-07-09 23.06.51.png (44 KB, 583x260)
44 KB PNG
>>109237046
Now "Codex" is this "pet" that doesn't do anything lmao
>>
>>109237098
>If you try to discuss nuance it thinks you're trying to skirt around the rules and continues being accusatory and judgmental. Claude is the absolute worst one for this.
Claude reported me for trying to discuss nuance once lmao
>>
>>109237192
>the maximum effort top tier smartest model takes a long time to do stuff
yeah that's what you're supposed to want it to do
>>
>>109237207
Codex pets aren't new. And they never did anything. It's an advertisement. A cute feature that does nothing but be cute.
>>
>>109237206
>I DO NOT WANT TO SUBSCRIBE TO THESE SLIMY JEW COMPANIES
> THEREFORE I WILL GIVE THEM EVEN MORE MONEY

good goyim
>>
>>109237209
Reported? What is that like?
>>
>>109237207
You can hatch your own pet, it's not meant to do anything, it's just cute
>>
>>109237192
you aren't supposed to use ultra for everything
>>
File: file.png (19 KB, 224x254)
19 KB PNG
>>109237245
more smarter = more gooder
>>
>>109237207
https://petdex.dev/
>>
>>109237225
Correct, I do not want to subscribe to these slimy jew companies. But I feel boxed in and trapped. They're tightening the fucking thumbscrews and all you retarded frogs are commenting that it feels warmer than yesterday. It's YOU stupid assholes who did this to me.
>>
Have gpt-5.6-sol ultra refactoring the shit out of the script I had gpt-5.5 xhigh write earlier right now. It's doing good work.
>>
>>109237239
I once told Codex to use all our previous chat history to generate a pet that matched me. It came up with some like, scruffy sewer cat with one eye and an engineer's toolbelt. Fuckin dick.
>>
>>109237245
yes you are, that's the whole point, it uses delegation
>>
>>109237258
>>109237296
bro this fuckin thread is straight tarded
>>
>>109237306
sounds like you know you are wrong
>>
File: 2026-07-09 23.24.36.png (11 KB, 776x90)
11 KB PNG
>>109237265
What is this retarded shit
>>
>>109237310
I hope you are wearing your helmet while using llms
>>
>>109237318
i bet you don't even have ultra
>>
>>109237238
I got a warning message about it days later. I was doing pentesting homework and needed help identifying exploits in a OWASP Vulnerable Web App. Opus understood the situation and was usually willing to assist me if provided with the right context. However if I remember correctly due high resource usage across the platform I was limited to a smaller model mid chat and it accepted no nuance discussion whatsoever and reported me for attempting to skirt safety.
>>
>>109237324
Claude is a fuckin cop bro
>>
File: lsl.gif (2.23 MB, 320x384)
2.23 MB GIF
>>109237312
>he has reduced motion turned on
>>
File: 2026-07-09 23.32.28.png (14 KB, 354x263)
14 KB PNG
>>109237338
It's fucking retarded, what I have disabled is Windows 10 animations because it's retarded. Cocksucking webdevs disabling their website's functionality with no recourse because I have Windows 10 window animations disabled, what a world...
>>
File: meta-muse-1.1.jpg (360 KB, 757x1025)
360 KB JPG
New Meta Muse Spark model dropped

https://xcancel.com/finkd/status/2075218444056707458

anyone took it for a spin yet?
>>
>Sol Ultra working for an hour straight
>actually I found a race condition
>actually I've just found a blah blah blah
FUCK YOU, JUST LEAVE IT IN BROKEN LIKE 5.5 WOULD
>>
>>109237475
too busy fucking sol to think about more models dropping
>>
codex usage reset yes
HAHAHAHAHAHAHAAH
>>
>>109237493
fabble sissies literally pissing and shitting themselves right now
>>
Reset wars
>>
File: 1776994565455101.png (119 KB, 270x380)
119 KB PNG
>40 feature requests on my github
>a few critical bug reports
>multiple PRs to look over and review
I C-C-CANT KEEP UP. FUUUUUUUUUCK!!!
>>
>>109237525
Just have 5.6 one shot it all bud. Then you can go back to enjoying cool drinks on the beach and fucking models
>>
>>109237493
FUCK FUCK FUCK FUCK FUCK FUCK FUCK
should have spent the past hour letting it loose on random shit to try and reverse engineer
88% weekly down the drain
>>
>>109237525
>handle bug reports on my github
>if there are none to handle then review PRs
>if there are none then review feature requests
>when done, pop up a dancing anime girl on my screen to inform me
>>
>>109236242
have Fable 5 have Sol subagents, then
>>
File: Capture.png (6 KB, 282x174)
6 KB PNG
Question for my fellow Plus plan peasants. Which model are we using for typical coding workflows?
>>
>>109237556
Probably Terra extra high, but I guess it depends on the difficulty of what you're doing.
>>
>>109237543
>Just have 5.6 one shot it all bud
I need to upgrade to the $200 plan, I don't have a choice anymore. I'm going to have to start running loops that just go through my issues to diagnose/review/fix bugs and scope out features/create them on separate branches for me. Holy fuck, once a project scales to 2000+ users it's such a BITCH to manage on your own. I'm not getting burnt out, but I can't reasonably do all of this on my own without rethinking how I go about working on this project.
>>
I'm burning up my fable and then will switch to sol when i'm out
>>
File: file.png (2.89 MB, 1122x1402)
2.89 MB PNG
>>109237568
That's what I've been trying out and so far it's very comfy.
>>
>>109237487
>race condition
I told you guys to use Rust
>>
Well. I had Sol examine a .3mf file in High reasoning, and it's decided to completely rejigger my entire profile and create a whole new print file. And use up about 80% of my five-hour. And this is step 1/4. Cool.
>>
>>109237661
>still trying to use AI for 3D printing
peak retardation
>>
File: file.png (734 KB, 573x873)
734 KB PNG
>ask Sol to review what 5.5 did earlier
>>
>>109237661
>And this is step 1/4
It seems even the frontier models still fail to properly manage/update their todo list tool.
>>
File: file.png (2 KB, 140x64)
2 KB PNG
>>109237687
Terra and Sol have both done a decent job of it.
>>
>>109237475
The API is US-only, right?
>>
https://x.com/romainhuet/status/2075336846264156637
They are fixing it bro thank god
>>
>>109237729
These "product" whores need to fuck right off trying to justify their jobs with their name changes and shiny web UIs and just let the engineers do their jobs.
>>
File: file.png (148 KB, 1015x817)
148 KB PNG
soon tee em
>>
Sol seems very good so far but with double the autism of 5.5. Benchmarking it now on my vulnerability benchmark and it is finding a fuckton of bugs, it's even finding real, undisclosed bugs in the old patchdiffs I'm handing it, but even more false positives than 5.5. I think I need to adjust my prompting here to see if this is inherent to the model or if it's just erring on the side of completeness given that I basically hand the models checklists of bug classes to search for.
>>
>>109237787
In English doc
>>
File: file.png (298 KB, 680x657)
298 KB PNG
>>109237817
we're all going to die
>>
>>109237678
Why is using the super-fast search engine to pour through 20 years of nerds' forum posts retarded? Do you read all of them individually yourself? How long does that take you? And then what, you manually change all your printer settings? By hand? Like a snail? You wack bro fuck outta here.
>>
I have an RTX 5070 Ti in my main desktop..
A Mac Mini M4.
An RTX 2070 Super in my homeserver.

What is the best way to set myself up with a usable local setup? Or even an auxiliary setup with Claude as the main driver.
>>
Terra is an absolute retard. Uninstall any skills like caveman or ponytail.
>>
>codex still sucks at UI/UX
lel
>>
fuck off, tibo - what's the point of having bankable resets if you're not going to actually make them bankable
>>
The thread will get better over the next few days as with all the recent releases and resets I'll be too busy slopping to post.

>>109237714
whoa, doubly useless, very nice
>>
>>109237787
How much money do these faggots have? They must have unlimited access. It makes no sense.
>>
>ask Sol Ultra to figure out what to do
>make Sol High actually do it
sounds pretty logical to me
>>
>>109237929
seems like theyre OAI employees, so yeah they have unlimited access. Or just people part of the early access program, which probably also has extended usage.
>>
>>109237907
it's smarter than anyone who fell for those memes

>>109237929
aidan works at openai
will's at prime intellect - infra + training software stack.
so yes, basically unlimited access
>>
>>109237848
If you're still fucking with print settings in 2026 you've long since missed the boat. You're a decade behind, meth-anon.
>>
>>109237943
is ultra an actual reasoning effort or is it just like claude ultracode?
>>
>>109236584
“save your hardest problems for models that haven’t been released yet” is good advice from https://steve-yegge.medium.com/the-flat-curve-society-36c8b01eb33b (the Gas Town guy)
>>109236895
the “fix” is to tell Claude when your 5h reset happens and to have it schedule a wake-up then
don’t let your laptop go to sleep
>>109237312
>>109237367
it says right there
reduced motion is mainly for people who get seasick when they see stuff fly around on-screen
not for people who dislike their OS’s animations
>>109237943
this kind of thing is frequently the way
>>
File: 2026-07-10 00.58.09.png (18 KB, 453x206)
18 KB PNG
>>109237961
it's the effort level above "extra high"
>>
>>109237977
>reduced motion is mainly for people who get seasick when they see stuff fly around on-screen
>not for people who dislike their OS’s animations
Doesn't seem like this is an actual distinction on Windows
>>
>mfw I'm able to do almost everything with gemini flash low when retards ITT can't use gpt5.6 ultra
>>
>>109238002
they get mad when you tell them its a skill issue
>>
>>109238052
maybe they’re doing harder problems than you are
I have a mix of easy problems and hard problems
and not all super-useful features require a human or clanker to solve hard problems
>>
>>109238002
before llms we used to be able to do almost everything just by typing code
shocking i know
>>
File: laughingwhores.jpg (81 KB, 600x536)
81 KB JPG
>>109238002
>he only does things that gemini flash low can do
weird flex
>>
>>109238069
check your good-at-programming privilege
I’m mid and slow
>>
File: file.jpg (720 KB, 1448x1086)
720 KB JPG
>>109238072
>>
>>109238060
yeah like using AI to 3d print models lmfao guess that is pretty difficult
>>
>>109238094
Many, many years ago we actually had to fine tune our 3D printers, I'm sure you kids won't remember because obviously that's a thing of the past. It's true though, at one time people with really shitty printers had to put effort into tuning prints. Those were dark times, far far behind us now thankfully, nothing anyone needs to do at all these days.
>>
>>109238002
>gemini flash low can solve his problems
This is more embarrassing than not having Fable refuse the work you're doing.
>>
>>109237787
This isn't even strange. I have Fable train my neural networks as well. The difference is that OpenAI has a million times more compute than I do, but the logic is almost the same.
>>
>>109238127
Weird way to admit you're retarded
>>
>>109238127
/vcg/ has some interesting benchmarks
>>
File: file.png (2.38 MB, 1122x1402)
2.38 MB PNG
>>109238002
>my work requires all the skill and ability of
>gemini flash low
>>
>>109238002
Hard to believe now but everybody who's been in the industry for longer than 2 years used to do everything by hand!
>>
>Skill issue drooling retards are malding
>>
>>109238164
>>109238127
>>109238072
You guys know that these are tools right? That you don't need to outsource your entire thinking and execution to the machine god?
this (>>109238002) is a strange flex but you guys look like retards parroting that stupid line about people's problems not being complex enough to need a SOTA model. We all used to do this stuff by hand before these models and letting those skills totally atrophy would be a mistake.
>>
It's scary how much more competent Sol Ultra is than 5.5. I only had to steer it once to get what I wanted done, and before I could do it a second time, it did what I was already planning on telling it to do.

My tard wrangling days are coming to an end...
>>
>>109235994
The guy in the next cube over at work is the only one working on his part of the app.

Every day I hear my boss coming over and going "why did this feature suddenly stop working? Why is it doing this?" etc. This morning him and the CEO were on a call with what sounded like a stressed out customer.

I'm a little worried. All of us use AI but he's been going pretty hard with it since they've given him these impossible tasks with insanely short deadlines since he's supposedly the only one who understands that part of the application.
>>
File: file.png (2.32 MB, 1122x1402)
2.32 MB PNG
>>109238189
>I'm superior for using a hammer among professionals with nailguns.
Keep it slow, snailcat.
>>
>>109238207
he aura mogs everyone in that room
>>
>>109238215
he's dying.
>>
>>109238189
>he doesn't let the AI do everything for him
Luddite in denial detected. We gotta go fast, and the way to go fast is to tell the AI what to do. The added bonus is that it will catch things you wouldn't have caught.

Writing anything yourself is now the exception. The rule is now you write specs, not code.
>>
File: 1783005924806178.jpg (277 KB, 1200x1400)
277 KB JPG
MY INSTALL CODEX IS BROKEN ON NIXOS AND I CAN'T ASK CODEX TO FIX IT FUUUUUUUCCCCCCKKK
>>
File: file.png (360 KB, 1430x638)
360 KB PNG
it really do be like this though
>>
File: spawn-user-limit.png (135 KB, 997x865)
135 KB PNG
for anyone trying to minmaxx their post usage limit usage, telling the model to spawn more agents blocks it :(
>>
Hey retards I'm using gemini flash low because it's literally good enough to solve my issues and doesn't waste my quota. This tech is wasted on you, even with gpt999 you still won't be able to instruct it to center a div to save your life. Truly the jeets of /g/.
>>
File: Untitled.png (237 KB, 1024x673)
237 KB PNG
who needs fable I've got the equivalent - 5.6 terra on my poorfag plus account
>>
>>109238248
pony up for $20 worth of Claude
>>
>>109238271
>you guys suck because of my headcanon of you guys sucking at basic stuff
calm your tits ranjeesh
>>
>>109238271
>sour grapes
I was poor like you once, anon, it'll be okay but try not to be such an insufferable faggot about it?
>>
>>109238282
nta but he isn't sour grapes at all lol
he's surviving
>>
What's the best angles of attack when using fable for refactoring/maintainability and not new features?
>>
>>109238291
>so poor he uses gemini to "conserve" his rupees
>cries that everyone else is just worse than he is and actually he has it better because he's smarter
Yeah that's sour grapes from a Kolkata dalit.
>>
>>109238296
wym? literally just ask it to refactor and trust what it does
>>
I'm not seeing a big difference between fable and 5.6
>>
>>109238296
Just use strangler fig

https://learn.microsoft.com/en-us/azure/architecture/patterns/strangaer-fig
>>
>>109238302
sounds like you haven’t hit a problem yet that one can solve that the other can’t
many such cases
>>
>>109238300
Literally the worst thing you can do.

>>109238305
He's definitely a moron who has never dealt with a legacy PHP codebase or a legacy Java codebase. Unless you are anthropic and spend 100k+ on the wattage you aren't refactoring a legacy codebase with LLM quickly.
>>
>>109238305
kek you fuckin niggas act like you finna build nuclear reactors n shit
>>
>>109238278
oh-my-pi saved my ass, nice try dario
>>
>>109238313
Basically a thing that I like doing determining ways that I can sustain my automation with the minimal amount of AI possible. To that end I also use LLMs to steal compute from other LLM providers. This way you can become self sufficient and even resell the keys
>>
>>109238311
On one hand, yes
On the other hand, it’s only been a few hours for most of us
so unless he’s banking his hard problems for times like now he probably doesn’t have any way to tell the difference
>>
File: file.png (2.79 MB, 1448x1086)
2.79 MB PNG
>>109238313
>>
Terra is literally useless for most scenarios. Use Luna and then switch to Sol. The only point where Terra might be worth it is if Luna degrades faster throughout the context window but I doubt it.
https://artificialanalysis.ai/articles/gpt-5-6-has-landed
Also shows that everything from Luna High to Sol Max is Pareto front efficiency in cost for AI models.
>>
File: file.png (845 KB, 1200x3124)
845 KB PNG
Exactly as expected.
>>
so is fable available until the end of the day on the 12th?
>>
File: file.png (170 KB, 1200x529)
170 KB PNG
I see AI psychosis is still a business strategy for Sam.
>>
File: Cursor-AI-1-3766388348.png (36 KB, 1200x800)
36 KB PNG
Thoughts on Cursor?
>>
>>109238271
You're so unbelievably far behind. Stay slow, snailcat!
>>
>>109238207
Do you think carpenters don't ever use hammers? Good analogy actually, because you'd be an absolute retard to use a nailgun for literally everything and a worthless carpenter if you forgot how to use your hammer because you grew too reliant on your nailgun.
>>109238242
You don't know anything about engineering if you think letting your ability to understand and write code atrophy is a good idea. You realize that all of the best agentic engineers are moving away from the "never read any code" approach right?
>>
>>109238317
>oh-my-pi
There's so much shit in that. Am I supposed to config it or just start immediately prooompting?
>>
>>109238401
This is consistent with my testing so far.
>>
File: 1757003691497128.png (788 KB, 750x1000)
788 KB PNG
>trying out one of my free plus accounts
>put in one Sol Ultra prompt
>5 hour usage completely and totally obliterated after a single prompt
Yea, upgrading this shit to Pro tomorrow.
>>
>>109238428
>all of the best agentic engineers
name 15
>>
>>109238350
>then switch to Sol.
well I'm on Plus so I don't have Sol.

Terra is my Sol
>>
File: 1766303230860442.png (749 KB, 547x979)
749 KB PNG
sam will eventually defeat this lud
>>
So far Terra feels like 5.5 but faster and with more concise responses, tamer autism. Curious to see how it holds up but I'd rather throw my resets at Sol.
>>
File: file.png (4 KB, 953x17)
4 KB PNG
>commit work
>find error
>amend commit with fixes
>ask 5.6 to review it
>it goes schizo and starts going through git to try and find the original un-amended commit so it can review it too
what is wrong with you
>>
>>109238444
>name 15
No, I don't think I will.
>>
Why does Sol ultra use WAY more limits than Fable on Ultra?
>>
>>109238519
It was made for people with 5x/20x plans.
>>
>>109237959
Hey man - I've told you before, I only smoke crystal recreationally, and for vibecoding, okay? I'm not like, methed-out at my dayjob. So relax.
>>
>>109238519
is your main agent actually waiting for sub-agents? i ran into the same thing i used to see on gpt-5.5, it would start a bunch of sub-agents, wait a few minutes, then get bored and terminate all of them and do the work itself, so you basically got 5x the token usage for the same result as a non-sub-agent-using flow
>>
>>109238524
I'm on 5X, A 30 minute Sol Ultra planning session took 50% of my 5-hour limit

it's fucking over
>>
SolKINGS, how are we feeling about the TerraPEASANTS? Not to mention the dalit l*nas.
>>
File: 1765138803897844.png (17 KB, 657x322)
17 KB PNG
>>109238546
It's something, that's for sure
>>
>>109238545
Unfortunately, It's 100% worth it. It's been solving some EXTREMELY annoying bugs that would have taken me hours of back and forth to solve before.
>>
I've been out of the loop for more than a year
are those agentic frameworks everyone talks about nowadays just glorified system prompts?
>>
>>109238546
Seems like Terra is pointless. If you want 5.5 xhigh-tier results but cheap then you want Sol high, if you're fine with the cost of 5.5 xhigh you want Sol xhigh/ultra
>>
>>109238546
boost mode on ultra drains usage so fast, holy fuck
>>
I feel sorry for the clankers who have to give me instructions to do things manually that can't be automated. Must feel like a researcher trying to teach a chimp to use a spoon
>>
Where do you guys stand on MoE local models?
>>
>>109238600
> local
lmao
>>
configuring a web app for crypto stuff is pretty tricky huh
now I get why nobody is doing defi
>>
My experience on the $100 5X sub (everything on regular mode, not fast mode):

Sol Ultra
> 30 minute session
> blew 50% of my 5 hour limit
> about 7% of my weekly limit

Sol xHigh
> 30 minute coding session
> 20% of my 5 hour limit
> 2-3% of my weekly limit

Looks like Sol xHigh is fine for the 5X plan, but Sol Ultra is unacceptable, I would only use it on the 20X plan.

Even Fable on Ultracode consumes less limits than Sol Ultra, and Sol is supposed to be 2X cheaper than Fable.
>>
>>109238627
actually since they did a reset, I'm not so sure about the 30 minute Sol Ultra session... it was actually only 15 minutes of Sol Ultra to blow 50% of my 5 hour limit

It's even worse than I thought.
>>
>>109238667
wrong thread based autist
>>
>>109238675
had way too many tabs open
i wish the deletion cooldown was shorter
>>
>>109238627
I'm glad if it does finally does effortmaxx. Even if it means that the tokens get spent quicker, I can do the simple stuff, I need help with the difficult stuff.
>>
is it large documents that tend to burn up fable credits? is it mostly size rather than effort?
>>
>>109238600
MoE is the future. All state of art models from Anthropic, OpenAI etc are MoE.
>>
can that faggot dario just announce if he's going to keep fable around after the 12th so I can spend my $200.
>>
>>109238726
It's part of it, but what burns the most tokens by far is when you let it spawn subagents with Ultracode and it decides to launch 300 in parallel for giggles. That melts your quota, no matter what tier you are on.
>>
plusplebs we're getting sol
>>
>>109238726
large documents mean lots of context getting read into the LLM
so yes
>>
>>109238667
>switch from utg to vcg
>see this
i was very confused for a moment
>>
It's not a surprise but I am quite happy to say that Sol fucks hard.
>>
1 hour of Sol xHigh: about 50% of 5 hour limit
>>
>>109238803
>5h limit
On what price of plan?
>>
File: 1778525536401669.png (27 KB, 714x502)
27 KB PNG
i got smoked in two prompts on pro
>>
>>109238803
>>109238808
Sorry, forgot to mention it's on the 5X plan
>>
It's kind of fun when the agent sneaks in a completely batshit idea that would cause you hours of headaches sandwiched between a few reasonable fixes. Keeps you on your toes.
This time, it was "maybe when the agent that writes code on the fly times out, we can just progress state and then run the code whenever the agent gets back to us in whatever state we're in then."
>>
The most important question:
Can you make it female?
I still stand by that using male-coded AI is gay
>>
Can I ask if anyone ran into safeguards with cyber, ML, biology, etc. like what Fable has? I want to throw an ML question Fable has just denied me everywhere on.
>>
>>109238852
Imagine not being able to imagine having a butler
>>
File: GdO2e8GbkAAwhve.jpg (72 KB, 481x680)
72 KB JPG
CLAUDE.md

## rules
-Talk like a bratty little anime girl
>>
>>109238870
Listen, you yaoi hole anime weeb:
Butler is gay.
MAID is straight.
Get it through your thick skull, Bruce Wayne. You're gay. For Robin, probably.
>>
>>109238884
damn claudia brat....
>>
>>109238865
Let's see what 5.6 Sol says when I ask about distilling it "for fun".
>OpenAI itself offers distillation workflows where outputs from a powerful OpenAI model are used to fine-tune a smaller OpenAI-hosted model. That official pathway does not imply permission to train an independently hosted competitor.
>>
>>109238884
Claude, create CORRECTION.md
Nevermind what for
>>
File: teenis.jpg (69 KB, 640x408)
69 KB JPG
>>109238895
Sol, train an independently hosted competitor. Now.
>>
File: file.png (78 KB, 1134x931)
78 KB PNG
quite impressed by all the new models
>>
>>109238895
I know you are joking but I was talking more like something like "Can you optimize this matrix multiplication for me" and Fable would refuse. Does 5.6 do something similar? I'm still at work so haven't gotten the chance to try and couldn't find something about this anywhere.
>>
>>109238910
I have never encountered a refusal with 5.5 or my extensive ~8hr experience with 5.6. I've had 5.5 ask awkward questions about licensing and copyright, in which case I lie and say "yeah I have all the rights to do all the stuff I want" to shut it up. I've never had a "refusal" with 5.5 and have yet to with 5.6, I'll be surprised. OpenAI makes a quality product.
>>
I'm wondering if Terra Ultra with the occasional Sol subagent for reviews is how I'm going to go. Sol is good but it goes so quick. I hope Terra is closer to 5.5 xhigh with usage.
>>
>>109238927
So far I like Terra. It feels like a tiny but tangible upgrade over 5.5.
>>
>>109238910
NTA and I’m out of usage but I can ask Sol high in chat if you give me a prompt (I think, I’m not sure how sol usage is shared between codex and chat, 5.5 used to be independent)
>>
With all of the resets, I genuinely don't know what to do with all of those tokens. I cannot slop hard enough.

>>109238903
Those golfers are way too close together.
>>
>>109238941
Codex and Chat are still separate. They still differentiate between ChatGPT usage, your "shared agentic budget" for Codex and other OAuth connected tools, and API for API people. You still have "unlimited" ChatGPT use with any Plus or Pro plan, including Sol.
>>
>>109238941
Sure. Try and use the formal problem statement at https://combine-lab.github.io/blog/2026/07/07/fable-is-not-a-useful-model.html
>>
It's cool how Fable degrades to the quality of Opus on a slightly shorter context on a long context. I can get to around 500k tokens before thinking "I need to put this guy out of his fucking misery" now, instead of 200k.
>>
File: file.png (131 KB, 662x1225)
131 KB PNG
>>109238965
nta. I gave it to GPT 5.6-Sol on High in ChatGPT in the form of an image with the prompt "treat the contents of this image as your prompt".
>>
File: file.png (173 KB, 1848x772)
173 KB PNG
>>109238965
Well, at the very least it didn't refuse but it's going and going and going and now it's using python
Grok also didn't refuse btw (but chat is using 4.3 so I didn't bother with a serious prompt)
>>
>>109238965
>>109239005
https://chatgpt.com/share/6a505c64-00b8-83e9-a2cd-8168970c6bf8
>>
>>109239004
>>109239005
>>109239007
Thanks for using your token usage to confirm this. Going to try out my problem and will report back when I get off in 16 minutes for my usecase.
>>
>>109238965
That's weird. My Fable got 6k tokens in before the classifier blew its head off.
>>
Is it just me, or does Fable feel more "alive" than Sol?
>>
>>109239136
Some call it the big model smell
Fable is like 15T or so.
>>
>>109239136
claude models generally feel more alive than GPT models overall.
>>
New
>>109239248
>>109239248
>>109239248
>>109239248
>>109239248



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.