Apple’s new SpeechAnalyzer API shows higher transcription accuracy than Whisper models and the legacy SFSpeechRecognizer while running faster on Apple hardware. The API replaced SFSpeechRecognizer with iOS 26 and macOS 26, and Apple has not released its own accuracy figures, so developers have been uncertain about migration. In tests run on an Apple M2 Pro using the LibriSpeech corpus, SpeechAnalyzer achieved a 2.12% word error rate on clean speech and 4.56% on noisy speech, compared with 9.02% and 16.25% for SFSpeechRecognizer and 3.74% and 7.95% for Whisper Small. SpeechAnalyzer transcribes roughly three times faster than Whisper Small and completes about 12 to 40 times more audio per second than real time. The benchmark used identical code paths for all engines, applied the same text normalization as OpenAI’s Whisper evaluation, and matched OpenAI’s published Whisper numbers, indicating reproducibility. The results are limited to English read speech on a single Apple Silicon machine; nevertheless, for English transcription on current iPhone or Mac devices the on-device SpeechAnalyzer engine now outperforms alternatives, and migration from SFSpeechRecognizer is recommended for better accuracy.