Pith. sign in

REVIEW 17 cited by

BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2411.13543 v2 pith:COB3EHCQ submitted 2024-11-20 cs.AI

classification cs.AI
keywords modelsagenticbalrogbenchmarkchallengingenvironmentsgamesllms
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large Language Models (LLMs) and Vision Language Models (VLMs) possess extensive knowledge and exhibit promising reasoning abilities, however, they still struggle to perform well in complex, dynamic environments. Real-world tasks require handling intricate interactions, advanced spatial reasoning, long-term planning, and continuous exploration of new strategies-areas in which we lack effective methodologies for comprehensively evaluating these capabilities. To address this gap, we introduce BALROG, a novel benchmark designed to assess the agentic capabilities of LLMs and VLMs through a diverse set of challenging games. Our benchmark incorporates a range of existing reinforcement learning environments with varying levels of difficulty, including tasks that are solvable by non-expert humans in seconds to extremely challenging ones that may take years to master (e.g., the NetHack Learning Environment). We devise fine-grained metrics to measure performance and conduct an extensive evaluation of several popular open-source and closed-source LLMs and VLMs. Our findings indicate that while current models achieve partial success in the easier games, they struggle significantly with more challenging tasks. Notably, we observe severe deficiencies in vision-based decision-making, as several models perform worse when visual representations of the environments are provided. We release BALROG as an open and user-friendly benchmark to facilitate future research and development in the agentic community. Code and Leaderboard at balrogai.com.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DungeonBench: A Benchmark for Rules-Rich Tactical Reasoning in Dungeons & Dragons Combat

    cs.AI 2026-07 conditional novelty 7.0 of 10

    DungeonBench scores LLM tactical play on D&D combat, finding frontier policies clear ~80% of single encounters but only 40% of linked multi-encounter days.

  2. Mastering PokeGym: Graph-Guided Multimodal Evolution at Test Time

    cs.CV 2026-04 unverdicted novelty 7.0 of 10

    The manuscript body introduces PokeGym, a vision-only automated 3D-game benchmark, while the abstract claims a G-EvoMAC method and 60.18% success rate absent from the body.

  3. LLM-WikiRace Benchmark: How Far Can LLMs Plan over Real-World Knowledge Graphs?

    cs.AI 2026-02 conditional novelty 6.0 of 10

    Frontier LLMs exceed human performance on easy Wikipedia navigation tasks but finish fewer than 25% of hard games, with failures driven by looping and an inability to replan after mistakes.

  4. Derivation and Numerical Simulation of a Thermodynamically Consistent Magneto Two-Phase Flow Model for Magnetic Drug Targeting

    math.NA 2025-08 conditional novelty 6.0 of 10

    Leading omni-modal AI models show superhuman memory but brittle cross-modal fusion: conflicting or redundant sensory input degrades performance, and removing a modality can sometimes improve it.

  5. LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra

    cs.MA 2025-07 reject novelty 6.0 of 10

    The LLM Economist framework couples persona-conditioned worker agents with an in-context RL planner to search US-bracket tax schedules, yet its Saez benchmark is derived from the planner's own solution and its headlin...

  6. StarDojo: Benchmarking Open-Ended Behaviors of Agentic Multimodal LLMs in Production-Living Simulations with Stardew Valley

    cs.AI 2025-07 conditional novelty 6.0 of 10

    StarDojo is a 1,000-task benchmark in Stardew Valley combining production and social activities, and the best tested MLLM (GPT-4.1) achieves only 12.7% success on its 100-task subset.

  7. MARBLE: A Hard Benchmark for Multimodal Spatial Reasoning and Planning

    cs.AI 2025-06 conditional novelty 6.0 of 10

    State-of-the-art multimodal language models perform at or near random chance on MARBLE, a new hard benchmark for spatial reasoning and planning.

  8. BYO-Eval: Build Your Own Dataset for Fine-Grained Visual Assessment of Multimodal Language Models

    cs.CV 2025-06 conditional novelty 6.0 of 10

    A Blender-based diagnostic toolkit that tests VLMs on fine-grained visual skills by varying one visual attribute at a time, exposing failure modes that coarse benchmarks miss.

  9. Enhancing Decision-Making of Large Language Models via Actor-Critic

    cs.CL 2025-06 conditional novelty 6.0 of 10

    LAC improves LLM decision-making by computing action scores from token logits and combining them with the model's prior policy through a gradient-free KL-constrained update.

  10. Sudoku-Bench: Evaluating creative reasoning with Sudoku variants

    cs.AI 2025-05 conditional novelty 6.0 of 10

    A new 100-puzzle Sudoku-variant benchmark is hard for frontier LLMs, which solve under 15 percent unaided.

  11. lmgame-Bench: How Good are LLMs at Playing Games?

    cs.AI 2025-05 conditional novelty 6.0 of 10

    lmgame-Bench turns six classic games into a scaffolded LLM evaluation suite, ranks 13 models, detects contamination, and reports RL transfer from Sokoban or Tetris to unseen games and planning tasks.

  12. Spatial Reasoning in LLM Game Agents: Impact of Causal Context and Multi-Step Planning

    cs.AI 2026-07 conditional novelty 5.0 of 10

    On a new three-game spatial benchmark, larger Qwen3 models with thinking mode and multi-step planning achieve higher win rates, while small models struggle to localize and causal prompt hints give only marginal, model...

  13. Pushing Forward Pareto Frontiers of Proactive Agents with Behavioral Agentic Optimization

    cs.AI 2026-02 conditional novelty 5.0 of 10

    BAO, a behavior-enhanced SFT plus regularized RL pipeline, improves proactive agents' task performance while lowering user-involvement rate, beating UserRL baselines on three UserRL gym tasks.

  14. TextAtari: 100K Frames Game Playing with Language Agents

    cs.CL 2025-06 conditional novelty 5.0 of 10

    TextAtari is a text-based Atari benchmark for language agents; 7-8B LLMs stay below 10% of human scores in over 90% of tested conditions, and knowledge injection helps more than chain-of-thought.

  15. Evaluation and Benchmarking of LLM Agents: A Survey

    cs.LG 2025-07 conditional novelty 4.0 of 10

    A review that proposes a two-dimensional taxonomy for evaluating LLM agents and highlights enterprise-specific evaluation gaps.

  16. General Modular Harness for LLM Agents in Multi-Turn Gaming Environments

    cs.AI 2025-07 conditional novelty 4.0 of 10

    A modular perception-memory-reasoning harness consistently improves LLM game scores across four puzzle games, with perception helping spatial games and memory helping long-horizon games.

  17. Evolutionary Perspectives on the Evaluation of LLM-Based AI Agents: A Comprehensive Survey

    cs.CL 2025-06 conditional novelty 4.0 of 10

    A survey that classifies AI agent evaluation benchmarks along environment and capability axes, and proposes five traits that distinguish agents from chatbots.

Pith tools