small models trained to specifically excel in one specific domain - currently only SFT is available. RL is running and will be ready soon