I've been testing GPT-5 for general reasoning and other tasks, and it’s not great there, but I haven’t tested it for code generation yet. From this eval, it looks like the model is marginally better. Definitely going to try it out. Thanks for the code, Akshay!