Voice model releases: OpenAI, Alibaba, and PolyAI

OpenAI released real-time and batch transcription models. Alibaba's Qwen-Audio-3.0-Realtime Plus tops speech-to-speech leaderboard (84.1% vs GPT's 79.1%). PolyAI released Dialog-RSN-1 audio-native voice model with sub-300ms latency, challenging speech-to-speech architectures. Concrete benchmarks and pricing available. Competitive shift in voice AI.