Microsoft MAI-Transcribe-2 Sets New Benchmark for Speech Recognition Accuracy

Microsoft's latest speech recognition model, MAI-Transcribe-2, has achieved near-top accuracy in transcription at a cost of $1.67 per 1,000 minutes.

This places it among the lowest-cost options in the high-accuracy tier, according to Artificial Analysis.

The model ranks second on accuracy among the models tracked by Artificial Analysis, with a 2.0% AA-WER (Average Words Error Rate).

MAI-Transcribe-2 is also notable for its throughput, which few competitors match.