Show HN: Kitchen Rush, Overcooked inspired LLM tool calling benchmark (github.com)

Kitchen Rush, a new benchmark tool for evaluating language models (LLMs), has been introduced to emphasize both accuracy and latency in tool-calling applications. Unlike existing benchmarks that mainly measure whether a model executes the correct function calls, Kitchen Rush combines these with time sensitivity, modeling a fast-paced cooking simulation reminiscent of the game Overcooked. In this environment, every decision influences real-time outcomes—orders must be fulfilled promptly to avoid penalties like expired requests, allowing researchers to see how well models perform under pressure. This benchmark is significant for the AI/ML community as it provides a clear metric, the Kitchen Rush score (KR), which quantifies a model's efficiency and effectiveness in real-time decision-making. By varying the latency budget for decision-making, Kitchen Rush enables the comparison of models tailored for different applications, such as voice assistants versus interactive agents. As speed is paramount for real-world deployments—where a model's ability to handle tasks promptly is as crucial as its accuracy—Kitchen Rush sets a new standard for evaluating the viability of LLMs in practical scenarios, showcasing how performance can drastically shift based on decision-making pace.