GoKawiilNari Labs' Qwen3-TTS and Qwen3-ASR top Coval's voice AI benchmark rankings
Nari Labs announced its Qwen3-TTS and Qwen3-ASR models achieved leading positions on Coval's voice AI benchmark, which measures latency and word error rate for speech AI systems. The ASR Fast model hit 44ms latency with 3.6% error rate at $0.12/hour, while the TTS Fast model achieved 63ms latency with a best-in-class 3.8% error rate, undercutting competitors like AssemblyAI and Deepgram on price.
GoKawiil's interpretation of the reporting above, not reported fact.
Voice AI agents depend heavily on low latency and accuracy to feel natural in real-time conversations, and Coval's benchmark is widely referenced by companies building these systems. Nari Labs' combination of competitive accuracy, speed, and dramatically lower pricing could pressure established players like AssemblyAI and Deepgram to compete more aggressively on cost for production voice applications.
- Nari Labs' Qwen3-ASR Fast model ranks #1 in latency (44ms) and #2 in accuracy (3.6% WER) among public STT models.
- Qwen3-TTS Fast ranks #1 in word error rate (3.8%) and #2 in latency (63ms) among public TTS models.
- Pricing undercuts rivals significantly, with ASR Fast costing $0.12/hour versus $0.45/hour for AssemblyAI's Universal 3.5 Pro.
Source: narilabs.com — Nari Labs Team, 2026-09-14
Published there as: “Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.