DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination
About
Multi-agent large language model (LLM) systems often fail to reliably outperform a single strong model equipped with best-of-N sampling. We argue that a core source of this instability is ill-posed equilibrium selection: current systems specify what information agents share, but not which coordination convention should be selected. We formalize a broad class of such systems as discounted incomplete-information Markov games and show that two common pathologies, oscillation between competing conventions and drift across them, can both induce unstable learning and linear Bayesian regret. To obtain a well-posed target, we introduce the Heterogeneous Quantal Response Equilibrium (HQRE), an entropy-regularized equilibrium concept with agent- and state-dependent temperatures. Under a monotonicity condition, HQRE is unique, admits linearly convergent mirror updates, and yields bounded Bayesian regret; the same condition yields rollout-measurable stability diagnostics. We instantiate this objective in two algorithms: DICE-PC, which coordinates frozen models through prompt-control actions, and DICE-FT, which performs parameter-efficient mirror fine-tuning. Across eleven benchmarks in four domains, DICE improves accuracy-cost trade-offs over strong within-class baselines; on reasoning and planning tasks, DICE-PC improves by 4.3 percentage points on average and DICE-FT by 8.5 points.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | MATH 500 | Accuracy99.1 | 589 | |
| Logic reasoning | ZebraLogic | Accuracy93.1 | 49 | |
| Travel Planning | TravelPlanner (test) | Final Pass Rate15.2 | 49 | |
| Planning | PlanBench | Accuracy42.7 | 47 | |
| Travel Planning | TravelPlanner (val) | Delivery Rate100 | 42 | |
| Mathematical Reasoning | AIME24 | Accuracy (AIME24)86.3 | 34 | |
| Planning | AutoLogic | Accuracy89.1 | 34 | |
| Multi-agent coordination | Overcooked-AI Mean of CR/AA/Ring/FC/CC layouts | Self-Play Score121.6 | 16 | |
| Multi-agent coordination | CollabEscape (test) | Success Rate80 | 13 | |
| Multi-agent coordination | CollabCapture (test) | Success Rate100 | 13 |