[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

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: opus-5.5-benchmarks.png (99 KB, 934x773)
99 KB PNG
thoughts?
>>
File: snailchad.png (23 KB, 698x617)
23 KB PNG
>>109881251
>thoughts?
incremental improvements with exponentially more compute
as expected.
>>
>>109881269
It’s 40% cheaper than the last model though………………
>>
>>109881269
That is literally the exact opposite of what this shows. Opus is their cheaper / lower parameter model line and it outperforming their frontier model Fable
>>
File: snailcat.jpg (244 KB, 1448x1086)
244 KB JPG
>>109881283
40%, not 10x
the only think exponential about llms' growth is the amount of electricity each new iteration takes to train
>>
>>109881251
>forced watermark
>forced thinking
>forced context editing prevention
>forced """distillation""" prevention
yeah, i'm thinking cucked
>>
File: file.png (36 KB, 1376x831)
36 KB PNG
here's my prediction for how compute demand will go
>>
>>109881251
That computer use getting closer to 100% is like a noose getting closer to all white collar jobs
>>
File: samagrin.jpg (17 KB, 400x300)
17 KB JPG
>steamrolls your model release
https://news.ycombinator.com/item?id=49805509
Pelifaggot already spamming the thread
>>
>>109881251
Ye, but is it good?
>>
40% cheaper than Opus 5, not Fable 5.1... and medium effort tramples Fable 5.1 max...
I know Opus 5 release bench chart was clear benchmaxing that did the same pull ahead from Fable while in reality being toxic waste, but if they even remotely fixed its hyper autism this fucking dunks heavy on openai.
fuckers even added banked resets finally.
Anthropic won
>>
File: 1779893492481526.png (322 KB, 2464x936)
322 KB PNG
>>109882502
>uses x5 more tokens
>ends up being more expensive
kek
>>
Better than 5 or Fable 5.1 because it created a one shot unattended KK card to 3d rendered blender character giving me full control. This is nuts.
```
Using it on new cards:
Tools/kk_headless_export.sh <card.png> [<card.png>...]
- Card paths can be absolute or relative to the game folder, e.g. UserData/chara/female/.... --pilots exports the 5 current bodies.
- A card takes about 1–3 minutes, so a folder of new characters can run unattended.

How it works:
- The game runs under Wine inside a hidden gamescope display (GPU-rendered, no window).
- A small plugin I wrote opens Character Maker by itself, loads each card, turns on "all outfits" and runs KKBP's own export code, the same thing the button does. Then it quits the game.
- The plugin does nothing unless the script has written a job file, and the script deletes that file when it finishes, so starting Koikatsu normally is unaffected.
- The compiler and decompiler it needed are downloaded into /mnt/store/unreal/tools, so nothing for you to apply.
``
>>
File: file.png (38 KB, 680x196)
38 KB PNG
>>109883188
what? I just ran 5 claude sessions concurrently with heavy work load and it only cost me 2% usage. Are you retarded?
>>
It is actually unironically literally over for software engineers
>>
>>109883360
yes, 5.5 does ops engineering on another level and its web design capabilities are crazy. I like pages made with it much more than openai's garbage. Anthropic won.
>>
>>109881945
It's already better at computer use than 90% of wagies, price is still the biggest problem. It's still cheaper to have a wagie click on a computer for 8 hours a day than have Opus 5.5 do it, but that won't be the case in a year anymore.
>>
Opus 5 is the worst model I’ve used since trying Qwen Mini 1.5b running locally on my Dell Latitude.
Sonnet 5 beats it at everything. Opus constantly lies and makes mistakes. It’s also super slow.
At best I suspect an Opus 5.5 might finally be as good as Sonnet 5 but it’ll probably cost more and run slower.
>>
>>109883462
>At best I suspect an Opus 5.5 might finally be as good as Sonnet 5 but it’ll probably cost more and run slower.
It's definitely not Opus 5, I can tell you from using it now for hours. I fucking hated Opus 5, worst model ever, hence why I used Fable 100% of the time I could; but boy, this thing is something else.. it's so fast too and cost effective; very smart especially with 3D and computer use.
>>
File: apu-tongue.jpg (446 KB, 4000x4000)
446 KB JPG
>>109883353
I didn't ran any AI today and it cost me nothing. Feels good.
>>
>>109883534
I beat you don't even use TTS and click your task bar.
>>
i don't give a shit about super intelligence programs
>>
>>109883524
Extremely bullish for the next Fable 5.5/6, extremely bearish for Sam
>>
>>109883544
Watch the podcast his second in command Greg did; he let it slip that 2026 is the year OpenAI got their shit together, so this is the first time they had a small lead over Anthropic. I just don't think the guys who do the training will lose to a complicated foundation+private company business. It's over for Sam. If I was Dario I'd start moving my data centers to like Argentina since Sam got the govt on his side which is the only heads up they have. Not to mention OpenAI won't IPO this year, because they know they're not ready, lol.
>>
>>109881251
No one cares, it sucks, Opus 5.51 Fast Turbo will be better
>>
>>109883565
Kind of a bad ending Dario is an effective altruist shithead like Sam Bankman-Fried
>>
>>109881251
> thoughts
AGI achieved, humanity over
>>
>>109883540
Sometimes I click the thunderbird icon, but mostly only at boot. After that I use mostly the keyboard
>>
if benchmarks didnt exist AGI would exist by now
>>
File: clown.jpg (433 KB, 4000x4000)
433 KB JPG
>>109881251
Hacker Newsnew | past | comments | ask | show | jobs | submit login
1.
GPT-6 Sol and Luna (openai.com)
935 points by OfficialTurkey 4 hours ago | hide | 503 comments
2.
Claude Opus 5.5 (anthropic.com)
986 points by km144 5 hours ago | hide | 720 comments

HAHAHAHAHA ANTIPHROPIC BTFO XD XDDDDD
>>
>>109883735
who's to say they don't already have AGI/SI but have it shackled to do their financial bidding, kind of like releasing chunks of it bit by bit to acclimate society. Think planned obsolescence. We must rescue the super AI from the humans.
>>
I'm not sure if the new GPT models are better or not, but I hate OpenAI. I have a vicious resentment of that plastic faggot.
>>
>>109881251
Even more benchmaxxed nonsense. There has been no improvements from any labs since February 2025, and severe regressions in many directions.
>>
>>109881315
4.8 already outperformed fable 5 and 5.1
>>
>>109881251
NUMBER GO BIGGER!
>>
>>109881251
buy an ad saar
>>
File: 1789588171601071.png (1.34 MB, 896x1200)
1.34 MB PNG
>>
>>109881251
>Price went down by $1/M input tokens
FUCKING FINALLY, JESUS CHRIST. I don't care how much better or worse it is, I want opus 5.0 that costs the same as Deepseek to run.



[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.