I'm a bit skeptical. This is just a single benchmark with an already capable (&closed-source) model. Would have been cooler if there was a more recent benchmark with an open-source model.