Just from a few trials, the output doesn’t seem to diverge much from the last “experimental”, although the newer version seems to wrap up open-ended listings of examples with fewer items.
Annoying is that the AI Studio site resets the temperature back to 1.0 = “unpredictable” when switching between models, so you have to be mindful both of this setting, and that low temperatures that produce close comparisons also lose some of the more creative language possibilities.