Nari Qwen3-TTS and Qwen3-ASR

Nari Labs serves open-source speech models behind production APIs, and the two in public beta are Nari Qwen3-TTS 1.7B for text-to-speech and Nari Qwen3-ASR 1.7B for speech-to-text. In Coval's voice AI benchmarks as of September 14, 2026, the company reports its Fast speech-to-text endpoint ranked first on time-to-final-segment at a p50 of 44 ms with a 3.6% word error rate, and its Fast text-to-speech endpoint first on word error rate at 3.8% with a 63 ms median time-to-first-audio. Coval's rankings can shift every 30 minutes and only publicly available endpoints were compared, so those positions are a snapshot rather than a fixed standing. Beyond the model APIs, Nari Labs offers dedicated inference for customers' own models and fine-tuning of multimodal models, and is backed by Y Combinator.