Comment by CompoundEyes

I’ve chipped on it with each model since 5.2 but 5.6 sol is something else. When it first came out I’d get some refusals but they’ve since stopped. I wonder what an ideal candidate benchmark task would be for that?