AssemblyAI puts 32 languages on realtime audio

AssemblyAI released Universal-3.6 Pro Realtime at $0.45 per audio hour, serving 32 languages with automatic language detection at the same latency as Universal-3.5 Pro. In its release benchmark, AssemblyAI says the comparison used 12,460 scripted caller scenarios with real speakers and matched hardware, while the rate card stays unchanged. For teams building voice interfaces beyond Mistral’s Voxtral transcription bet, this is a reminder to test noisy rooms, accents and phone lines—not only clean English—before paying for a second speech stack.