DPBench: Structural Determinants of Multi-Agent LLM Coordination

Researchers introduced DPBench, a benchmark evaluating coordination in multi-agent LLM systems, finding that protocol structure—not model capability—determines deadlock rates. GPT-5.2 achieved 25% deadlock while Gemini 2.5 Flash reached 90%, but protocol changes like pre-commitment communication or resource-ordering prompts eliminated deadlock entirely.

Computer Science Artificial Intelligence Submitted on 2 Feb 2026 v1 https://arxiv.org/abs/2602.13255v1 , last revised 3 Jun 2026 this version, v2 Title:DPBench: Structural Determinants of Multi-Agent LLM Coordination Under Simultaneous Resource Contention View PDF /pdf/2602.13255 HTML experimental https://arxiv.org/html/2602.13255v2 Abstract:We present DPBench, a benchmark for evaluating coordination in multi-agent systems built from large language models. Existing benchmarks measure task-level success under a fixed protocol; the structural conditions under which coordination succeeds or fails at all have not been characterised. DPBench adapts the Dining Philosophers problem into a controlled testbed where the action protocol, the communication structure, and the group size each vary independently. We evaluate six agents: GPT-5.2, Claude Opus 4.5, Grok 4.1, Gemini 2.5 Flash, Llama 4 Maverick, and a uniform-random baseline. Under simultaneous action at N=5 with the default prompt, deadlock ranges from 25.0% 95% Wilson CI 11.2, 46.9 for GPT-5.2 to 90.0% 74.4, 96.5 for Gemini 2.5 Flash; sequential action is solved by four of the six. Holding the model fixed at Gemini 2.5 Flash, three protocol variables drive deadlock from 90% to within CI of zero: three rounds of pre-commitment communication 0.0% vs. single-round 86.7% , a prompt encoding a classical concurrency primitive 0.0% for resource-ordering and symmetry-breaking, against 100% for the minimal prompt , or doubling the group from N=5 to N=10 90.0% to 10.0% . Single-round messaging and memory of past timesteps do not change the rate at the sample size we ran. Whether the same model coordinates or deadlocks is determined by the protocol, not by the model's capability. Submission history From: Prashanth BusiReddyGari view email /show-email/ffd0ccfd/2602.13255 Mon, 2 Feb 2026 18:26:00 UTC 65 KB v1 /abs/2602.13255v1 v2 Wed, 3 Jun 2026 20:03:36 UTC 184 KB References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs https://info.arxiv.org/labs/index.html .