The author compared Mica v0.1 4B and Laya (ModernBERT‑large 421M) on an identical Tetris sequence using llama.cpp, with both models producing zero output tokens per move. Each turn a heuristic narrows the options to four placements, which are shuffled; Mica selects by reading the probability of its answer‑label tokens, while Laya scores the options via its encoder. The experiment used three seeds on a 10×20 board, invoking one model call per move.