Microsoft's new speech model is 10x faster and crazy cheap
Microsoft just launched MAI-Transcribe-2, claiming the crown for the fastest, most accurate, and cheapest speech model available. It hits a 5.2% Word-Error-Rate across 60 languages while running up to 10x faster than OpenAI's GPT-Transcribe. It also packs speaker diarization and word-level timestamps at a promo price of $0.10 per hour through year-end.