The most prestigious law school admissions discussion board in the world.
|
 |
 
|
GPT 5.6 in the regular chat window seems smarter than Fable 5
Poast new message in this thread
Date: August 20th, 2026 1:14 PM Author: The Penis Not just in codex, obviously codex sol x high or ultra is going to be smarter than a regular fable chat. But I mean like just the regular chat gpt window on the web app that thinks for 3 seconds seems smarter than running the fable api on max reasoning max tokens. open ai is obviously doing something different that doesn't immediately translate to the metrics that are being used to evaluate these things (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080552) |
 |
Date: August 20th, 2026 1:37 PM Author: TurboGrafx-67 Not just in codex, obviously codex sol x high or ultra is going to be smarter than a regular fable chat.
explain |
|
Date: August 20th, 2026 1:41 PM Author: The Penis You can't see why an agent process that runs for 1-2 hours and spawns multiple subagents to delegate various tasks including creating files, running tests and code, and using tools and skills is going to be better than a chat window that thinks for 2 min and 12 seconds? (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080616) |
|
Date: August 20th, 2026 1:43 PM Author: TurboGrafx-67 Wait what? Both fable and opus 5 can use dynamic workflows and I've seen them auto orchestrate subagents without even requesting
Maybe try using Cowork one time. Claude chat is pointless for actual work now unless your job involves navel gazing in the weeds about abstract problems |
|
Date: August 20th, 2026 1:49 PM Author: The Penis I didn't say fable 5 COULDNT use agents, I said obviously codex is better than fable in a regular chat session
"Claude chat is pointless for actual work now unless your job involves navel gazing in the weeds about abstract problems"
--that's my point. that's why I said it's obvious. I meant noone would contest that codex is going to be better than a regular chat response. |
|
Date: August 20th, 2026 1:53 PM Author: TurboGrafx-67 Codex shouldn't be compared against Claude chat lol
My point was that cowork is a closer comparison. Try it in the desktop or iPhone app |
|
Date: August 20th, 2026 1:56 PM Author: The Penis Right that was my point. I was saying " obviously codex sol x high or ultra is going to be smarter than a regular fable chat.", as in that's obvious, everyone knows that is comparing apples to oranges. But that even just comparing gpt 5.6 on high reasoning in the regular web app it still is smarter than Fable in the api on max reasoning max tokens, despite that Fable is the highest performing chat model in terms of stats. (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080664) |
|
Date: August 20th, 2026 1:59 PM Author: TurboGrafx-67 We are talking past each other
We both agree Fable is generally sps
Just curious what your source was or if you had a published benchmark because everything I saw put Fable at the top which gives me tremendous cognitive dissonance |
Date: August 20th, 2026 1:22 PM Author: cannon Claude went downhill fast (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080562) |
|
Date: August 20th, 2026 1:23 PM Author: agenticman (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080567) |
|
Date: August 20th, 2026 1:36 PM Author: coalition of rabbis i never got why you guys liked it so much i tried both and codex was instantly better imo (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080595) |
|
Date: August 20th, 2026 1:36 PM Author: literal females on a jury in court (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080597) |
|
Date: August 20th, 2026 1:37 PM Author: The Penis I have always liked gpt best. Only anthropic model I have ever liked is Fable, but it still seems overrated. (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080601) |
|
Date: August 20th, 2026 1:37 PM Author: TurboGrafx-67 They peaked with Opus 4.6 max extended reasoning (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080605) |
|
Date: August 20th, 2026 1:39 PM Author: The Penis Nah opus 4.6 sucks now that I'm used to gpt 5.6 (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080608) |
|
Date: August 20th, 2026 1:40 PM Author: TurboGrafx-67 They nerfed it but I use 5.6 in chat/work/codex and opus 4.6 finds errors daily in gpt work (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080610) |
|
Date: August 20th, 2026 1:45 PM Author: The Penis It's probably all hallucinated. It finds "errors" that don't actually make a difference and creates endless nonsense "work". I have gone back a ton and used opus 4.6 in aws that hasn't altered the original weights since release, across various subjects. It has way less of a knowledge base, makes tons of reasoning errors, hallucinates slop a fuckton, is overly sycophantic, overly confident about weak conclusions etc. It is nowhere even close to gpt 5.6. Even when opus 4.6 first came out I prefered gpt 5.2 or whatever the equivalent was back then to it. (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080631) |
|
Date: August 20th, 2026 1:49 PM Author: TurboGrafx-67 No it finds basic errors in doc drafting that my paralegal wouldn't even make
Or mistakes of fact or law (more rare)
No ai is flawless and the biggest mistake anyone can make is depending on one model
But you're right that fable and opus 5 max are horrible at review and create more issues than they solve |
|
Date: August 20th, 2026 1:41 PM Author: agenticman cr that model was building out full end to end apps/modules without stopping. this current iteration with human guardrails/and managing tokens "Say the word and I'll do it" is a nightmare to work with.
|
|
Date: August 20th, 2026 1:53 PM Author: Oh, You Ricky? what is TCR model to use for creating custom data point XML type spreadhseet info into a grpahical display and then generating a report
|
|
Date: August 20th, 2026 3:01 PM Author: agenticman prob Sol 5.6 on codex (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080825) |
|
Date: August 20th, 2026 1:23 PM Author: .- .-. . .-. . .--. - .. .-.. . Claude is still the only game in town for manipulating office docs. I want to have sex with the PowerPoints it creates. (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080565) |
|
Date: August 20th, 2026 1:36 PM Author: coalition of rabbis not even close to being true (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080598) |
|
Date: August 20th, 2026 1:27 PM Author: literal females on a jury in court Anthropic is heavily throttling their models
They're going full scam mode |
|
Date: August 20th, 2026 1:33 PM Author: .- .-. . .-. . .--. - .. .-.. . The thing burned $500 of tokens in two days for me but papered an entire private markets programs for me from scratch so well that the biglaw firm I sent it to for review couldnt find a single thing to edit and just harped about the need to review executed sub docs well and "controls" around QP/AI status like a bunch of tards when that shit was always a given.
The thing was smart enough to even find a gap in my form CoFs that would have precluded bestowing clean title in shares. This was the one and only docs I drafted myself and it all but called me a retard not fit to breath air on the same planet as itself.
It not only papered all the legal docs but the project plan it built for my ops folks was better than anything a human could do without spending a solid week on the one Excel.
Its better than any human right fucking now. Its only going to get scarier in the future.
|
|
Date: August 20th, 2026 1:37 PM Author: coalition of rabbis change the speed to standard or slow
no reason to ever use fast imo |
|
Date: August 20th, 2026 1:38 PM Author: The Penis reason to use fast is that if you give it a hard problem it takes like 2 hours to run. But also I guess that's nbd anyway. gives you moar time to poast. (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080606) |
|
Date: August 20th, 2026 1:38 PM Author: coalition of rabbis ^ (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080607) |
|
Date: August 20th, 2026 1:40 PM Author: Taylor Swift is not a hobby she is a lifestyle (πΊπΈ π΅π±) (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080612) |
|
Date: August 20th, 2026 1:41 PM Author: .- .-. . .-. . .--. - .. .-.. . People bitching about speed are just mega spoiled brats. If you sent the same shit to outside counsel you would wait a week for docs back. Now you wait 2 hours -- oh, the humanity! (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080618) |
|
Date: August 20th, 2026 1:42 PM Author: coalition of rabbis (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080620) |
|
Date: August 20th, 2026 1:51 PM Author: Oh, You Ricky? how mcuh more token does it burn
i reallyl dont want to use Claude but it seems like the best model for somethign i need vut i dont want to burn money |
|
Date: August 20th, 2026 1:41 PM Author: TurboGrafx-67 I like slow responses. Makes me feel like the ai tried harder (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080619) |
|
Date: August 20th, 2026 1:48 PM Author: The Penis I just like that it wastes more time (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080638) |
|
Date: August 20th, 2026 1:53 PM Author: coalition of rabbis (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080652) |
|
Date: August 20th, 2026 1:47 PM Author: literal females on a jury in court https://x.com/GoySuperstar/status/2090291660819513641 (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080636) |
|
Date: August 20th, 2026 1:54 PM Author: coalition of rabbis what the fuck is that common? literally only used claude for one day until i was like this is inferior (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080658) |
|
Date: August 20th, 2026 1:57 PM Author: literal females on a jury in court I'm with you man, exact same experience here. He is right that the kind of people who use Claude are lonely weirdos who want a companion rather than a tool (http://www.autoadmit.com/thread.php?thread_id=5895259&forum_id=2",#50080667) |