The coding abilities are many times better, there are fewer coding mistakes but the speed is significantly slower and stopping tasks due errors is very frequent.
During my last 15 tasks (4 versions each), this is what I keep seeing:
6 times: “Failed to sample tokens” - longest after 32 minutes.
9 times: “You don’t have the ability to clone this repository” - longest after 1h 0m
3 times: “An unknown error occurred” - 1h 11m
Finally, it does it’s job but fails to produce any code (4 times):
