GPT-5.4 benchmarking better than Claude Opus 4.6 on computer use, web browsing, knowledge work tasks and agents tool use 👀 Time to try it out