75 ms P50 time-to-first-audio on Speko; ranked 33rd of 112 in the TTS for Voice Agents 2026 index.
Benchmark summary: 14.6/100, ranked 33rd of 112 in the TTS for Voice Agents 2026 index (Listed). Strongest criterion: price, 2.4 of 5 points.
Nari Qwen3-TTS is an open-weights text-to-speech (TTS) API from Alibaba (Qwen), served by Nari Labs, measured by one independent benchmark, Speko.
Overall score
14.6 /100
Data checked on September 30, 2026
Updated after each benchmark capture
Every point comes from a public benchmark: sources · methodology
Nari Qwen3-TTS ranks 33rd of 112 in the 2026 text-to-speech (TTS) API benchmark, with a score of 14.6/100.
Speko measures a P50 TTFA of 75 ms from US East (n=30), dated September 26, 2026: 3rd lowest of the 29 models Speko measures.
Speko rates its pronunciation robustness on numbers, dates and currency at 0.32 on a 0-to-1 scale (13th highest of the 28 models Speko rates).
Speko rates its naturalness at 1,510 Elo from blind A/B votes (23rd highest of the 41 models Speko rates).
Speko lists a cost of about $5 per 1M characters.
Speko measures this model through Nari Labs' API, so the latency above is Nari Labs'. Speko lists it under Nari Labs; the creator, Alibaba (Qwen), is inferred from the model name.
Speko changed its pronunciation robustness between the captures of September 30, 2026 and October 1, 2026 without a new measurement date (0.9, then 0.32): the score uses the average, 0.61. The value shown is the latest.
Google Chirp 3: HD is a proprietary text-to-speech (TTS) API from Google, measured by two independent benchmarks: Artificial Analysis and Coval.
Deepgram Aura 2 En is a proprietary text-to-speech (TTS) API from Deepgram, served by Cloudflare, measured by one independent benchmark, Coval.
OpenAI GPT-4o mini TTS is a proprietary text-to-speech (TTS) API from OpenAI, measured by two independent benchmarks: Coval and Speko.