Hugging Face researchers have benchmarked frontier automatic speech recognition (ASR) systems against code-switched speech—the linguistic phenomenon where bilingual speakers mix languages within conversations. The analysis reveals significant performance gaps when customers seamlessly blend languages, a common occurrence in multilingual customer service environments. The findings highlight critical vulnerabilities in current voice agent technology that could impact real-world deployment in diverse markets. The study evaluated leading ASR models' ability to accurately transcribe speech that alternates between languages mid-sentence or mid-word, scenarios that are linguistically natural but computationally challenging. Code-switching presents unique obstacles because it requires models trained on multiple languages simultaneously while maintaining contextual awareness across language boundaries. These benchmarks provide essential insights for companies developing voice agents targeting bilingual and multilingual populations.