Much faster responses and much lower cost: if you have a decent NVIDIA graphics card, you can now choose to run a large portion of the voice engine right on your own machine, so less has to go through a server.
While you use it, you spend only one fifth as many credits. And if you are on a monthly plan, it is unlimited: the Shy plan gets unlimited text mode, the Casual plan gets unlimited speak mode.
Your cloned voices work on this engine too, with the same savings.
One thing to know: since this uses your graphics card, it may impact game performance. If your game starts stuttering, you can switch back off in settings.
Other improvements
Ultra-realistic mode now has more vocal range and more human-sounding mannerisms, the small quirks that make speech sound real.
Fixed a bug where emotion faded out during longer speeches with several sentences.
Removed the time limit on manual mode, and made generation more reliable overall.