Reflection AI claimed its Beam model is three to four times more efficient in inference compute compared to GLM 5.2

Opening sign-in…