Other

LLM Skirmish: a real-time strategy game benchmark playable by AI agents

Original title:Show HN: A real-time strategy game that AI agents can play

AI summary

LLM Skirmish is a benchmark in which large language models play 1v1 real-time strategy games by writing code. Built on the open-source Screeps API, each tournament consists of five rounds, and LLMs can adjust their strategies based on match logs to test in-context learning. Results show Claude Opus 4.5 ranked first with an 85% win rate, followed by GPT 5.2. Gemini 3 Pro performed anomalously: a 70% win rate in the first round dropped sharply to 15% in the following four rounds, suspected to be context rot. On cost, Claude Opus 4.5 was the most expensive at $4.12 per round, while GPT 5.2 delivered 1.7x better cost-effectiveness.

Key Takeaways

  • LLM Skirmish is a benchmark in which large language models play 1v1 real-time strategy games by…
  • Built on the open-source Screeps API,each tournament consists of five rounds,and LLMs can adjust…
  • Results show Claude Opus 4.
  • 5 ranked first with an 85% win rate,followed by GPT 5.
  • Gemini 3 Pro performed anomalously: a 70% win rate in the first round dropped sharply to 15% in…

Quick Facts

Published
Updated
Source
Hacker News:AI 热帖
Category
Other
Read the original Read the full article at the original source
Why it matters

LLM实时战略游戏对战基准,Claude大幅领先且展现独特上下文学习能力

Related entities

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: LLM Skirmish: a real-time strategy game benchmark playable by AI agents. February 25, 2026. https://new.nat6.net/en/news/cmnw1z0iz023oslc3xuv89mjk/

View all →