Toga III 0.3. Net 0.6, the first net trained from reinforcement learning. Maybe 20 elo stronger in self play.