Benchmarks place GPT 5.5 as the best model on SWE, but is it the best at making apps end-to-end? Turns out Opus 4.8 continues to be the king of vibe coding on both price & performance. Introducing ViBench: the first benchmark for app creation based on real world tasks