LMGame‑Leaderboard

Leaderboard ↗

Built on LMGame‑Bench, we maintain two cutting‑edge leaderboards:

1. Model Leaderboard – ranks 19 state‑of‑the‑art LLM/VLMs evaluated without the agentic harness.
2. Agent Leaderboard – ranks 14 state‑of‑the‑art LLM/VLMs with our perception‑memory‑reasoning harness enabled.

Both tables cover gameplay across Super Mario Bros, Sokoban, 2048, Candy Crush, Tetris, and Ace Attorney.