AsyncTool: Evaluating the Asynchronous Function Calling Capability under Multi-Task Scenarios

Researchers introduced AsyncTool, a benchmark for evaluating asynchronous tool calling in LLM agents.

The benchmark simulates concurrent, heterogeneous tasks with realistic tool-response latency across step, sub-task, and task levels.

Results show that delayed tool responses can significantly degrade performance in multitask settings.

The study underscores the need for better coordination, dependency tracking, and temporal reasoning in LLM agents.