SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models

In three linesSMAC-Talk extends StarCraft Multi-Agent Challenge with natural language communication to evaluate LLM-based agents in cooperative multi-agent environments. Open-source benchmark testing decentralized control, partial observability and long-horizon decision-making, including scenarios with deceptive communicators. Evaluation on Qwen3.5 models.

Summary generated by Claude — human-verified