Topic
multi-agent
Multi-Agent RL System MAMO Automates Weight Selection for Constrained Optimization Problems
MAMO decouples task execution from objective design using multi-agent RL to automatically select reward weights for constrained optimization, improving adaptability in dynamic environments.
Before the Pull Request: Mining Multi-Agent Coordination to Solve the Trust Gap in AI Coding Agents
A study on multi-agent coordination in AI coding agents reveals that monitoring pre-PR coordination can reduce duplicate work to near zero and triple useful throughput. The open-source tool grite records agent interactions inside git, enabling recovery of failure modes invisible in pull-request history.
MEAL Benchmark Enables Continuous Multi-Agent RL Training on 100 Tasks in Hours Using GPU Acceleration
Researchers introduced MEAL (Multi-agent Environments for Adaptive Learning), the first benchmark for continual multi-agent reinforcement learning. Using JAX and GPU acceleration, MEAL enables training on sequences of 100 tasks in hours on a single GPU, revealing failure modes not apparent at smaller scales. This addresses the limitation of previous benchmarks that only considered 3-10 sequential tasks due to CPU constraints.
New Multi-Agent AI Pipeline Delivers Auditable Financial Chart QA with On-Premise Deployment
A new multi-agent AI pipeline called AgentFinVQA enables auditable financial chart question answering with on-premise deployment, improving accuracy by up to 7.68 percentage points over baselines and providing a verifier-based confidence signal for human-in-the-loop review. The system records every step in a traceable Model Evaluation Packet.
Hidden Anchors Reveal Why Multi-Agent LLM Deliberation Escapes Groupthink
A new paper from arXiv models multi-agent LLM deliberation as a closed-loop dynamical system where each agent has a hidden internal belief, or anchor, that continually pulls its opinion. The model explains how agents' confidence can climb past where any agent started, escaping the convex hull of initial beliefs. Tests across three open-weight model families show the anchor's influence is a spectrum.
WorkflowPerturb Benchmark Offers Calibrated Stress Tests for Multi-Agent Workflow Metrics
WorkflowPerturb introduces 4,973 golden workflows and 44,757 perturbed variants with three perturbation types at severity levels 10%, 30%, and 50%, enabling calibrated interpretation of workflow evaluation metrics for change management in multi-agent systems.
MedCollab Multi-Agent Framework Outperforms Leading LLMs in Clinical Diagnosis Accuracy
A new multi-agent framework called MedCollab, guided by Issue-Based Information System (IBIS) protocol and Hierarchical Disease Relation Chains (HDRC), demonstrates superior performance in clinical diagnosis and report generation. Tests on ClinicalBench and MIMIC-IV show it outperforms leading LLMs and medical multi-agent baselines in accuracy, evidence consistency, and reasoning quality.
Multi-Agent Peer-Reviewed Reasoning Boosts LLM Accuracy in Medical Question Answering
Researchers designed a multi-agent peer-reviewed reasoning method for medical question answering, where multiple LLMs generate and evaluate each other's chain-of-thought reasoning. Experiments with five models on three benchmarks showed the approach consistently outperforms single-model reasoning and majority voting, achieving best accuracy of 0.820. The method scales effectively and improves interpretability.
LectūraAgents Multi-Agent Framework Promises Adaptive Personalized AI-Assisted Learning
Researchers propose LectūraAgents, a multi-agent framework for adaptive personalized AI-assisted learning. It uses a hierarchical architecture with a ProfessorAgent leading specialized agents to generate and deliver tailored lecture content with embodied teaching actions. The system was validated on diverse courses and showed gains in content quality and personalization.
Mind-Studio: Executable World Models with Lookahead Evaluation for Partially Observable Games
Researchers present Mind-Studio, a framework that uses large language models to synthesize executable world models from state-action-next-state trajectories. On Montezuma's Revenge, it improves next-state prediction from 0.3% to 48.7% and verifies 5 of 8 subgoals, outperforming prior approaches.
XFlow: A New Programming System for Reliable Multi-Agent Workflows Addresses Prompt–Harness Boundary
Researchers present XFlow, an executable protocol programming system designed to improve reliability in LLM-based multi-agent workflows. By introducing the XPF protocol language and lifecycle-governed symbols, XFlow makes constraints and process requirements explicit and enforceable, addressing the underspecified prompt–harness boundary that limits current systems.
CoffeeBench: New Benchmark Evaluates LLM Agents in Multi-Agent Economic Simulations
Researchers introduce CoffeeBench, a benchmark for evaluating LLM agents in a long-horizon multi-agent economy. The 90-day simulation features farmers, roasters, and retailers, with models controlling one roaster. All models outperformed a passive baseline, but Claude Haiku 4.5 showed an idle-drift failure mode.
Tensor-Coord: Algebraic Decomposition Enables Conflict-Free Multi-Agent LLM Planning
A new research paper introduces Tensor-Coord, a multilinear algebra framework that represents joint plans of multiple LLM agents as a third-order tensor. By decomposing the tensor, it identifies coordination conflicts and enables iterative replanning, achieving 100% conflict-free plans for 2-agent tasks and 80% for 3-agent tasks in simulated delivery scenarios.
AdaSTORM Breakthrough Scales LLM Reasoning to Thousand-Node Dynamic Graphs, Paves Way for Supply Chain AI
AdaSTORM, a new multi-agent AI framework, scales large language model reasoning to dynamic graphs of up to thousand nodes with over 90% accuracy. The approach uses adaptive partitioning and collaborative reasoning to overcome limitations of current LLMs, which can only handle tens of nodes. This breakthrough could enable AI-driven analysis of complex, evolving networks such as supply chains.
Researchers Propose QoS-Aware Token Scheduling and Private Data Valuation for Multi-Modal Agentic Networks
A new arXiv paper introduces a QoS-aware token scheduling and private data valuation framework for decentralized multi-modal agentic networks. The approach embeds multi-modal data in a shared semantic space and uses differentially private prototypes to balance utility and privacy, showing improved fairness and QoS in simulations.
StateGen Platform Generates Synthetic Training Data for Tool-Augmented LLMs with 9.66/10 Hallucination Score
Researchers introduce StateGen, a synthetic data generation platform that produces scored, reasoning-trace-rich training conversations for tool-augmented LLMs. The platform uses a four-role LLM loop and an authoritative state manager to eliminate tool-call hallucinations, achieving a 9.66/10 score across 64,698 evaluated conversations.
AgentLeak Benchmark Reveals Internal Channel Privacy Leaks in Multi-Agent LLM Systems
A new benchmark called AgentLeak evaluates privacy leakage in multi-agent large language model (LLM) systems, finding that inter-agent messages leak at 68.8% compared to 27.2% for final outputs. Across 1,000 scenarios and five models, total system exposure reaches 68.9%, highlighting risks invisible to standard output-only audits.
Synthetic Counteradaptation: A New Framework for Human-AI Co-evolution in Enterprise Systems
A new research paper introduces synthetic counteradaptation, a principle describing how humans and AI systems co-evolve by adapting to each other's strategies. The paper analyzes examples from the game of Go, mixed-motive social interactions, and geopolitical simulations, providing a framework for understanding recursive human-AI dynamics in multi-agent environments.