Research map / Large language models / Agents, code and math
lmgame-Bench: How Good are LLMs at Playing Games?
Lanxiang Hu, Mingjia Huo, Yuxuan Zhang, Haoyang Yu, Eric P. Xing, Ion Stoica, Tajana Rosing, Haojian Jin, Hao Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add perception and memory harness | o3 · Candy Crush | score: 106.0 -> 647.0 | Improved |
| add perception and memory harness | o3 · Super Mario Bros | score: 1955.0 -> 3445.0 | Improved |
| add harness | claude-3-5-sonnet · Super Mario Bros | score: 1540.0 -> 1267.7 | Hurt |
| add memory module (ablation) | Llama-4-Maverick · 2048 | score: 44.6 -> 98.1 | Improved |
| add vision (symbolic) module (ablation) | o4-mini · Sokoban | score: 1.3 -> 5.3 | Improved |
| add memory module (ablation) | o4-mini · Sokoban | score: 1.3 -> 1.3 | No clear effect |
| add memory module (ablation) | Gemini-2.5-Pro · 2048 | score: 120.5 -> 118.0 | No clear effect |
| train Qwen2.5-7B-Instruct with RL on Sokoban | Qwen2.5-7B-Instruct · Sokoban 6x6 | success rate: 11.3 -> 24.2 | Improved |
| train with RL on Sokoban | Qwen2.5-7B-Instruct · Blocksworld 2D | success rate: 9.0 -> 29.5 | Improved |
| train with RL on Sokoban | Qwen2.5-7B-Instruct · WebShop | success rate: 7.0 -> 19.1 | Improved |
| train with RL on Sokoban | Qwen2.5-7B-Instruct · BIRD | success rate: 25.0 -> 17.5 | Hurt |
| train with RL on Sokoban | Qwen2.5-7B-Instruct · GSM8K 1 turn | success rate: 89.5 -> 89.0 | No clear effect |
Closest papers by meaning
- GameArena: Evaluating LLM Reasoning through Live Computer Games (ICLR 2025 · similarity 0.90)
- Are Large Vision Language Models Good Game Players? (ICLR 2025 · similarity 0.90)
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (ICLR 2025 · similarity 0.89)
- GameVerse: Can Vision-Language Models Learn from Video-based Reflection? (ICML 2026 · similarity 0.88)
- GlitchBench: Can Large Multimodal Models Detect Video Game Glitches? (CVPR 2024 · similarity 0.88)
- KORGym: A Dynamic Game Platform for LLM Reasoning Evaluation (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
