“Google DeepMind's smaller "flash" models are reportedly 95% as capable as its frontier models while costing only one-tenth the price to run.”