The MIQ Benchmark: Evaluating Machine Intelligence Quotient in Open-Ended Agentic Environments

ABSTRACT

As large language models transition from passive dialogue systems to active, autonomous agents, traditional benchmarks evaluating static question-answering capabilities have become obsolete. We introduce the Machine Intelligence Quotient (MIQ) benchmark, a comprehensive evaluation framework designed for open-ended agentic environments. MIQ assesses agents across five core dimensions: Long-Horizon Reasoning, Tool Proficiency, Authority Delegation, Contextual Memory Retention, and Safety Compliance. By deploying agents in highly dynamic, multi-step simulated web and enterprise environments, we demonstrate that models excelling in standard NLP tasks often fail dramatically at agentic orchestration. The MIQ framework provides a rigorous, reproducible methodology for scoring the true operational reliability of autonomous swarms.