Apple's on-device speech recognition API, SpeechAnalyzer, introduced for iOS 26 and macOS Tahoe 26, has surpassed OpenAI's Whisper Small model in English benchmarks, according to a comparison published by the development team behind the on-device AI app 'Inscribe'. The team compared SpeechAnalyzer with OpenAI's Whisper models (Tiny, Base, and Small) using 2,620 clear reading audios and 2,939 more difficult and noisy reading audios from the 'LibriSpeech' dataset. The evaluation metric was 'WER,' which indicates the word error rate. Apple announced SpeechAnalyzer at its Worldwide Developers Conference (WWDC25) in June 2025, describing it as the next generation of its speech recognition API that supports long recordings, conversations, and audio recorded from remote locations. Apple's official announcement did not include specific figures showing how accurate SpeechAnalyzer is compared to conventional methods or other companies' speech recognition models, prompting the Inscribe team to conduct the benchmark.