AppWorld

The 'AppWorld' dataset/benchmark contains a collection of applications and their associated environments, used to evaluate the performance of language agents in understanding and executing user instructions within complex contexts.

Papers using AppWorld (15)

From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents2026From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws2026When Should Agent Trust Be Conditional? Characterizing and Attacking Skill-Conditional Reputation in Agent Swarms2026ACCORD: Action-Conditioned Contextual Grounding for Language Agents2026Hera: Learning Long-Horizon Coordination for Device-Cloud Collaborative LLM Agents2026ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents2026Pushing the Limits of LLM Tool Calling via Experiential Knowledge Integration and Activation2026HINT-SD: Targeted Hindsight Self-Distillation for Long-Horizon Agents2026Keep Policy Gradient in Charge: Sibling-Guided Credit Distillation for Long-Horizon Tool-Use Agents2026Three Roles, One Model: Role Orchestration At Inference Time To Close The Performance Gap Between Small And Large Agents2026CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution2026Toward Scalable Verifiable Reward: Proxy State-Based Evaluation for Multi-turn Tool-Calling LLM Agents2026Reinforcement Learning for Self-Improving Agent with Skill Library2025ACON: Optimizing Context Compression for Long-horizon LLM Agents2025Prost: Progressive Sub-task Training For Pareto-optimal Multi-agent Systems Using Small Language Models2025