Nari Labs has announced that its voice AI models are leading the Coval voice AI benchmarks for both Text-to-Speech (TTS) and Speech-to-Text (STT) as of mid-September 2026. Coval provides industry-standard evaluation for speech AI agents, measuring critical metrics such as Word Error Rate (WER) and latency.

In the STT benchmark, the Qwen3-ASR Fast model ranked #1 in Time-to-Final-Segment (TTFS), with a median of 44 ms and a WER of 3.6%. In the TTS benchmark, the Qwen3-TTS Fast model ranked #2 in Time-to-First-Audio (TTFA) at 63 ms and achieved #1 in WER with 3.8%. Nari Labs noted that while the model shares the same architecture as the official Qwen3 TTS Flash Realtime endpoint, its service provides significantly lower latency and higher accuracy.

The company also highlighted the cost-efficiency of its endpoints. The Qwen3-TTS Fast model is tied for the lowest price on Coval's pricing directory at $10 per 1 million characters. The Qwen3-ASR Fast model is tied for the second-lowest price among models with public rates at $0.12 per hour.

Sources: