A new benchmark, RuneBench, tests large language models' ability to play the game RuneScape by providing them with the game's wiki. The benchmark reveals that some models, including Claude, overthink strategies and spend excessive time reading documentation, which negatively impacts their performance. Other models like Gemini and GPT also show suboptimal performance in specific in-game tasks. AI
IMPACT Highlights how current LLMs may struggle with complex, real-world tasks requiring strategic planning and execution.
RANK_REASON The item discusses a benchmark of LLMs playing a video game, which is commentary on LLM capabilities rather than a core AI release or research.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →