The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Codex 5.3 extra-high is so good its nuts
| robot daddy | 02/06/26 |
| .,.,.;;,;.,;:,:,,:,.,:,::,..;.,:,.:;.:.,;.:.,:.::, | 02/06/26 |
| robot daddy | 02/06/26 |
| .,.,...,..,.,.,:,,:,...,:::,...,:,.,.:..:. | 02/06/26 |
Poast new message in this thread
Date: February 6th, 2026 12:25 AM Author: robot daddy also way faster than 5.2 (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2Reputation#49650246) |
Date: February 6th, 2026 12:25 AM Author: .,.,.;;,;.,;:,:,,:,.,:,::,..;.,:,.:;.:.,;.:.,:.::, what do you use it for (http://www.autoadmit.com/thread.php?thread_id=5831622&forum_id=2Reputation#49650249) |
 |
Date: February 6th, 2026 12:31 AM Author: robot daddy 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=2Reputation#49650256) |
|
Date: February 6th, 2026 1:37 AM Author: .,.,...,..,.,.,:,,:,...,:::,...,:,.,.:..:. 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=2Reputation#49650280) |