400k context with 100% on the fiction livebench would make GPT-5 the undisputably best model IMHO. Don't think it will achieve that though, sadly.