“The R1-Zero model exhibits reliability issues, such as changing languages during reasoning.”