FaceXBench introduces the first systematic evaluation of Multimodal Large Language Models (MLLMs) on complex face understanding capabilities, critical for security applications.

• Comprehensive benchmark with 5,000 multimodal questions derived from 25 public datasets
• Evaluates MLLMs on face recognition, authentication, and analysis tasks
• Reveals current limitations and capabilities of leading models in biometric security contexts

This research has significant security implications by establishing how well AI systems can understand facial features used in authentication systems, potentially identifying vulnerabilities in biometric security protocols.

FaceXBench: Evaluating Multimodal LLMs on Face Understanding