This is grok, and it ■■■■■. Even Claude 3.5 makes less failures
1 Like
tried supernova and it blew me away in how bad it is.
back to gpt-5 
2 Likes
It seems to cheat a lot to solve the problem for example a testcase failed and I asked it to fix the test, it just deleted the testcase data which made the test work but not actually test anything 
6 Likes
I tried it on and off over the last week. I don’t think it ever did what I hoped for even once. Terrible. I thought it was going to be an upgraded grok-code-fast-1 but it is slower and takes strange actions, like triggering the tool that thinks I asked to edit a file that is outside of the project during a suuuuuuper simple refactor request. I switched to grok-code-fast-1and said “continue” and it finished the tasks perfectly in an eye-blink.
I’m now using primarily GPT-5 (high) and sometimes medium, and grok-code-fast-1 for simple fast things and rarely using sonnet except to occasionally code review something that is already done or simply something over-engineered by GPT-5. …but this could change again any week, as most of us probably know.
3 Likes
agree with this, i am now always using grok-code-fast-1 if need speed and if need visual and more complex using GPT-5 / GPT-5-Mini
code-supernova ignore rule, slow and miss with my prompt that grok or gpt 5 can understand
Be careful! this one actually deletes your files without permission! (Tested for a few days.) I think it might be Elon’s AI, it’s terrible at coding and following instructions. Back to Claude, time is money!
1 Like
Supernova gives a feeling of Claude 3.5
2 Likes