Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost

Nari Labs has open-sourced Qwen3-TTS and Qwen3-ASR, two high-accuracy, low-latency speech models, which have been shown to outperform closed-source models in benchmarks. The models are available on GitHub and have been designed to be fast and cheap to run, with the Qwen3-TTS endpoint achieving sub-50 ms latency at 10 requests per second (RPS).