The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Codex 5.3 extra-high is so good its nuts
| Avocado community account principal's office | 02/06/26 |
| Exhilarant Stage | 02/06/26 |
| Avocado community account principal's office | 02/06/26 |
| spruce comical abode crotch | 02/06/26 |
| hateful tanning salon | 02/06/26 |
| robot daddy | 02/10/26 |
Poast new message in this thread
Date: February 6th, 2026 12:25 AM Author: Avocado community account principal's office also way faster than 5.2 (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2\u0026mark_id=5301927#49650246) |
Date: February 6th, 2026 12:25 AM Author: Exhilarant Stage what do you use it for (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2\u0026mark_id=5301927#49650249) |
 |
Date: February 6th, 2026 12:31 AM Author: Avocado community account principal's office AI-related development. I just used it on extra high mode and it edited (and created some new unit tests) 14 files in one shot, and fixed it to a quality level that would have taken like 12 rewrites with codex 5.2. plus it was like twice as fast. (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2\u0026mark_id=5301927#49650256) |
|
Date: February 6th, 2026 1:37 AM Author: spruce comical abode crotch The SWE Bench Pro graph with number of tokens vs. accuracy is interesting. At the limit, it converges on 5.2 codex max. Models are now becoming more token efficient, but not necessarily developing new capabilities? (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2\u0026mark_id=5301927#49650280) |
|
Date: February 6th, 2026 1:48 AM Author: hateful tanning salon I got into the alpha for the codex app
This thing is amazing |
|
Date: February 10th, 2026 2:04 PM Author: robot daddy its so 180. I have heard people complain that its an electron app and that developers won't use it. but whats funny is I normally do almost everything headless but now that this app is out i just want to do everything on here. and speed and scale aren't really an issue. its fast enough and in terms of scale the credits are enough if I wanted to add more compute and more complex automation it would just get out of control and I'd have no idea what the hell is going on in my codebase. the credits on pro are enough. i could probably push it past the limits but i just would lose track of whats happening (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2\u0026mark_id=5301927#49661179) |