Benchmarking Open-Ended Multi-Agent Coordination in Language Agents

2026-06-06

AI Summary

Current evaluation frameworks struggle to assess the comprehensive capabilities of language agents in open-ended, long-horizon, multi-agent collaborative tasks. This work proposes ALEM—a benchmark environment built on the Craftax dynamics and implemented in JAX—that uniquely integrates openness, temporal extension, and multi-agent collaboration. By incorporating procedurally generated tasks, controllable collaboration difficulty, explicit communication mechanisms, and memory-augmented reasoning modules, ALEM cleanly disentangles individual competence from collaborative ability and provides a quantifiable testbed. Experiments reveal that 13 state-of-the-art LLMs achieve only 6% average normalized return under zero-shot settings. Notably, Gemini-3.1-Pro-High approaches the performance of billion-step-trained MARL agents in the most challenging scenarios, whereas GPT-5.4-High, despite high task scores, exhibits weak collaboration—demonstrating that collaborative proficiency is distinct from individual capability.

Abstract

As language models are increasingly deployed as autonomous agents, they must coordinate with others over long horizons in open-ended interactive tasks. Yet existing evaluations rarely test these demands together, instead emphasising single-agent tasks, short interactions, or highly structured multi-agent settings. We introduce $alem$, a JAX-based benchmark for open-ended multi-agent coordination built on Craftax-like dynamics. Alem embeds procedurally generated coordination tasks, soft specialisation, communication, and controllable coordination difficulty into a long-horizon survival world with exploration, crafting, trading, and combat. We evaluate $13$ modern LLMs zero-shot within homogeneous teams, with trained MARL agents as reference points. Current LLM agents remain far from solving alem, averaging only ~6% normalised return, but their failures are not uniform. On the hardest coordination setting, zero-shot Gemini-3.1-Pro-High approaches MARL agents trained for one billion steps, while GPT-5.4-High achieves strong base-task reward but much lower coordination reward. This contrast shows that individual task competence does not imply coordination competence. Ablations show that communication is the largest contributor to coordination, while memory and reasoning help when used to maintain multi-step plans. Overall, our results identify coordination as a distinct bottleneck for frontier LLM agents, separate from single-agent capabilities. Alem makes this bottleneck measurable and provides a controlled testbed for developing agents that communicate, allocate roles, and execute shared plans. Code is available at https://github.com/alem-world/alem-env.

Problem

• multi-agent coordination
• open-ended tasks
• language agents
• long-horizon interaction
• benchmarking

Innovation

• multi-agent coordination
• open-ended tasks
• language agents
• procedural task generation
• long-horizon interaction