The thing I find most baffling about the programming tests I've been running is that tools based on the same large language model tend to perform quite differently. Also: The best AI for coding in ...