Have tried all and continue to eval regularly. I spend up to 14 hours a day. Currently recovering from a herniated disk because I spent 6 weeks sitting at a dining room table, 14 hours a day, leaning foward. Don't do that. lol. So my coverage is pretty good. I'm using GPT5-codex-high for 99% of my work. Also I have a team of 40 folks, about a third of which are software engineers and the other third are cybersecurity analysts, so I get feedback from them too and we go deep on our engineering calls re the latest learnings and capabilities.