Humanness Index™ · TTS model

Neuphonic neu_hq

neu_hq is the quality tier of Neuphonic's text to speech line and the model name its SDKs and integrations carry by default.

Rank: #18
Humanness: 41
Likely rank: #17–19
Blind votes: 1,045

Standings as of Jul 28, 2026, 02:12 UTC

A real arena clip: a cloned source voice reading a customer support prompt at phone quality.

neu_hq key stats

Latency (measured): 276 ms¹
Languages: 9²
Streaming: Yes³
Voice cloning: instant, ~6 s sample⁴
Released: August 2025⁵

Vapi streaming benchmark (50 trials per model) (checked 2026-06-11) Median of 50 sequential live streaming trials, June 2026; includes network RTT from the benchmark machine.
docs.neuphonic.com/build-group/languages (checked 2026-06-12) English, Chinese, French, German, Japanese, Korean, Spanish, Portuguese, and Urdu.
docs.neuphonic.com/build-group/text-to-speech (checked 2026-06-12) SSE and WebSocket streaming surfaces.
docs.neuphonic.com/build-group/voice-cloning (checked 2026-06-12) Docs ask for a 6 second or longer clean sample; in practice the clone endpoint accepts roughly 3-10 s clips.
docs.neuphonic.com/changelog (checked 2026-06-12, confidence: medium) Neuphonic published no model-specific date; its changelog marks API v1.0.0 in August 2025. The neu_hq name predates GA (the closed beta opened in late 2024).

Background

neu_hq is the quality tier of Neuphonic's text to speech line and the model name its SDKs and integrations carry by default. Neuphonic, a London startup founded in 2024, pairs the hosted API with an open source on-device family (NeuTTS), and the hosted service streams over SSE and WebSockets across nine languages with instant voice cloning.

Sources: docs.neuphonic.com, github.com

At a glance

Neuphonic's public API currently serves a single voice pool and does not expose model selection, so this entry represents that served system under its neu_hq branding. The historical neu_fast realtime sibling is not listed separately: with one pool behind the API, a second row would test the same model twice. In our 50 trial streaming benchmark it returned first audio in a median of 276 ms including network time.

Sources: docs.neuphonic.com

Position in the rankings

Standings as of Jul 28, 2026, 02:12 UTC

Rank	Provider	Model	Humanness	Latency
#16	Cartesia	Sonic 3	65	166 ms
#17	Smallest.ai	Lightning v3.1	45	420 ms
#18	Neuphonic	neu_hq	41	276 ms
#19	Gradium	Gradium TTS	35	332 ms
#20	Cartesia	Sonic	0	116 ms

See the full Humanness Index™ rankings

Frequently asked questions

How is neu_hq tested on the Humanness Index™?: Listeners hear neu_hq against another model in a blind head to head round, both voices reading the same customer support prompt from the same cloned source voice, and they pick whichever sounds more human. Its Humanness score derives purely from those votes.
Why is neu_fast not on the Index?: Neuphonic historically exposed neu_fast as the realtime sibling of neu_hq, but the current public API does not expose model selection and serves one voice pool. The Index lists that pool once, under the neu_hq name; a separate neu_fast entry would put two names on the same system.

Keep exploring

NeuphonicAll Neuphonic models on the Index

Back to the Humanness Index™

Find the most human-sounding voice for your agent.

Compare the models in blind tests, read the methodology, or get in touch.

Read the methodology Star on GitHub

Build a TTS model? Add yours to the Index.

neu_hq key stats

Latency (measured)

276 ms¹

Languages

9²

Streaming

Yes³

Voice cloning

instant, ~6 s sample⁴

Released

August 2025⁵

Vapi streaming benchmark (50 trials per model) (checked 2026-06-11) Median of 50 sequential live streaming trials, June 2026; includes network RTT from the benchmark machine.

docs.neuphonic.com/build-group/languages (checked 2026-06-12) English, Chinese, French, German, Japanese, Korean, Spanish, Portuguese, and Urdu.

docs.neuphonic.com/build-group/text-to-speech (checked 2026-06-12) SSE and WebSocket streaming surfaces.

docs.neuphonic.com/build-group/voice-cloning (checked 2026-06-12) Docs ask for a 6 second or longer clean sample; in practice the clone endpoint accepts roughly 3-10 s clips.

docs.neuphonic.com/changelog (checked 2026-06-12, confidence: medium) Neuphonic published no model-specific date; its changelog marks API v1.0.0 in August 2025. The neu_hq name predates GA (the closed beta opened in late 2024).

Background

At a glance

Rank

Provider

Model

Humanness

Latency

#16

Cartesia

Sonic 3

166 ms

#17

Smallest.ai

Lightning v3.1

420 ms

#18

Neuphonic

neu_hq

276 ms

#19

Gradium

Gradium TTS

332 ms

#20

Cartesia

Sonic

116 ms

Frequently asked questions

How is neu_hq tested on the Humanness Index™?

Listeners hear neu_hq against another model in a blind head to head round, both voices reading the same customer support prompt from the same cloned source voice, and they pick whichever sounds more human. Its Humanness score derives purely from those votes.

Why is neu_fast not on the Index?

Neuphonic historically exposed neu_fast as the realtime sibling of neu_hq, but the current public API does not expose model selection and serves one voice pool. The Index lists that pool once, under the neu_hq name; a separate neu_fast entry would put two names on the same system.