The most prestigious law school admissions discussion board in the world.
|
 |
 
|
oh noes tp, what do you think of CODEX SOL?
Poast new message in this thread
Date: July 18th, 2026 2:17 PM Author: The Penis IMO Fable is strongest of all models for one-shot prompt questions, but codex sol ultra literally destroys it. I guess it isn't a fair comparison to judge a chat session to a long running multi-agent process, but also codex 5.5 beat fable 5 apples-to-apples in an agent benchmark in a huge upset, so it's not just that. Assuming codex 5.6 beats fable even more in fable-agent runs. (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008385) |
Date: July 18th, 2026 2:19 PM Author: Jared Baumeister https://x.com/i/status/2078370395426541775 (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008391) |
 |
Date: July 18th, 2026 2:21 PM Author: The Penis lol some dude's subjective opinion. also kimi uses twice as many tokens so the price can't be compared directly like that. (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008394) |
|
Date: July 18th, 2026 2:21 PM Author: gaetan dugas I haven't used anything past 3.4
Not understanding what people need beyond that |
|
Date: July 18th, 2026 2:24 PM Author: The Penis 5.6 ultra is 180 because you can get it to spawn subagents right in the basic prompt interface without having to do any additional scripting or use an orchestrator or anything (actually think I have seen it do it in xhigh too). I use prompts where I have everything pass through like 3 gates of stages where it fails if it doesn't make it and spawns new subagents each time (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008410) |
|
Date: July 18th, 2026 2:24 PM Author: gunnerretttired (gunneratttt) What do you use for legal and other office work? Cowork? I've just used my company's AI internal stuff and Grok for personal chatbot stuff, but I wanna make the plunge into something more substantive for personal stuff. (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008407) |
|
Date: July 18th, 2026 2:30 PM Author: The Penis Right now been using a combination of the Fable api and codex sol. I use Fable for brainstorming and testing ideas aggressively before wasting a codex ultra run, which can take up to 2 hours. You can actually use codex for just about anything, it isn't just a developer tool. Open AI even combined everything into the one codex app. But they also have a "work" section now that I think is supposed to be similar to Claude's cowork tool. Haven't used the work section yet, because I'm used to codex. Another thing about codex is that it has access to various "skills" that are useful for particular tasks. For instance it has a skill where it will use computer vision to do a visual inspection of a pdf to check the formatting when it is finished. If you want something nicely typeset you can have it write it in latex and it comes out perfect at the end due to this tool, because it corrects the code if it looks like shit. And yeah, it isn't just for developing, for instance in your prompt you can delegate subagents to run concurrent research tasks and lit reviews on whatever subject you want. (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008425) |
|
Date: July 18th, 2026 2:37 PM Author: gunnerretttired (gunneratttt) Tyft! (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008441) |
|
Date: July 18th, 2026 2:38 PM Author: have you ever looked at a quarter multi-threaded Aquinas lit reviews tp (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008444) |
|
Date: July 18th, 2026 2:40 PM Author: gunnerretttired (gunneratttt) Artificial intelligence using multi-threaded agents to sync your god-image (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008448) |
|
Date: July 18th, 2026 2:41 PM Author: have you ever looked at a quarter (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008449) |
|
Date: July 18th, 2026 2:34 PM Author: Nude Karlstack (🧐) GPT models have always been superior. Periodically an anthropic model is better on coding specifically, but for general use it has never been close
But everything is so good now that a slight edge in coding is basically irrelevant, where the general capability gap is still quite noticable. Plus to the extent all AI writes like shit Anthropic models especially always yap too much. Last time I used fable in an agent it went buckwild adding paragraph long comments with nonsense jargon |
|
Date: July 18th, 2026 2:35 PM Author: have you ever looked at a quarter /*Usurping the quandary*/ (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008435) |
|
Date: July 18th, 2026 2:35 PM Author: Nude Karlstack (🧐) Jfc (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008436) |
|
Date: July 18th, 2026 2:41 PM Author: The Penis I'm considering writing a custom multi-agent orchestrator for Opus through bedrock, because I have money on my aws account to burn. Do you use the anthropic agent sdk for fable agents or do you use graph lang or something custom? (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008452) |
|
Date: July 18th, 2026 2:43 PM Author: have you ever looked at a quarter what could you possibly be doing (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008454) |
|
Date: July 18th, 2026 2:43 PM Author: The Penis metaphysics (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008457) |
|
Date: July 18th, 2026 2:45 PM Author: have you ever looked at a quarter i had to subscribe to chat gpt to get this (1 month free):
A person ensures the actualization of their nature by freely choosing the actions that perfect that nature. That's the answer Aquinas gives. Everything beyond that belongs to his broader metaphysics of primary and secondary causality. |
|
Date: July 18th, 2026 2:48 PM Author: Nude Karlstack (🧐) I use only pi with a straight $200 codex sub. When I play around with fable or other new US models I am directly using my company's bedrock provider through pi instead
I don't do anything complicated enough to need orchestration (not counting stuff like when a model will spin up sub agents to look up multiple things at once). I make crud app go brrrr 🤤
Same setup on my personal computer. On the rare occasion I need something I just ask pi to make itself an extension (like I have a little module purely for system configuration).
The only downside to this setup is terminal UIs suck ass. |
|
Date: July 18th, 2026 2:36 PM Author: Mega Penis ITT: fat virgin nerds (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008438) |
|
Date: July 18th, 2026 2:36 PM Author: Nude Karlstack (🧐) You're the guy who has middle age sex like some faggot (http://www.autoadmit.com/thread.php?thread_id=5883883&forum_id=2/en-en#50008439) |