Humanness Index™ · TTS model

Lightning v3.1

Lightning v3.1 is Smallest.ai's flagship text to speech model, a 44.1 kHz engine built for conversational agents with instant voice cloning from a 5 to 15 second sample.

Rank: #14
Humanness: 62
Likely rank: #1–20
Blind votes: 3

Standings as of Jun 13, 2026, 02:28 UTC

LowerHigher

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

Lightning v3.1 key stats

Latency (measured): 420 ms¹
Languages: 12²
Price / 1M chars: $15³
Streaming: Yes⁴
Voice cloning: instant, 5-15 s sample⁵
Released: January 2026⁶

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.smallest.ai/waves/model-cards/text-to-speech/lightning-v-3-1 (checked 2026-06-12) Model card: 12 languages with auto-detection and code-switching; the launch post lists 15 including major European languages.
smallest.ai/pricing (checked 2026-06-12) Lightning V3.1 pay-as-you-go at ~$0.0145 per 1k characters = $14.50 per 1M; the Lightning V3.1 Pro tier bills ~$0.0195 per 1k ($19.50 per 1M).
docs.smallest.ai/waves/model-cards/text-to-speech/lightning-v-3-1 (checked 2026-06-12) HTTP, SSE, and WebSocket surfaces on the unified Waves API.
docs.smallest.ai/waves/documentation/voice-cloning/instant-clone-api (checked 2026-06-12)
smallestai-ff1e543d.mintlify.app/v4.0.0/content/changelog/announcements (checked 2026-06-12, confidence: medium) Waves changelog lists the v3.1 release in January 2026; the Lightning V3 family launch post followed on 2026-03-27.

Background

Lightning v3.1 is Smallest.ai's flagship text to speech model, a 44.1 kHz engine built for conversational agents with instant voice cloning from a 5 to 15 second sample. It reads 12 languages with automatic language detection and mid-sentence code-switching, and its launch benchmark published head to head listener wins on EmergentTTS against GPT-4o-mini TTS, ElevenLabs Turbo v2.5 and Multilingual v2, and Cartesia Sonic-3.

Sources: docs.smallest.ai, smallest.ai

At a glance

The first Smallest.ai entry on the Index, served through the unified Waves API with HTTP, SSE, and WebSocket access. Pay-as-you-go pricing sits at roughly $14.50 per 1M characters, among the lowest rates in the field. In our 50 trial streaming benchmark it returned first audio in a median of 420 ms including network time.

Sources: smallest.ai, docs.smallest.ai

Position in the rankings

Standings as of Jun 13, 2026, 02:28 UTC

Rank	Provider	Model	Humanness	Latency
#12	ElevenLabs	Turbo v2	64	302 ms
#13	Neuphonic	neu_hq	63	276 ms
#14	Smallest.ai	Lightning v3.1	62	420 ms
#15	MiniMax	Speech 2 HD	62	357 ms
#16	ElevenLabs	Multilingual v2	59	1006 ms

See the full Humanness Index™ rankings

Frequently asked questions

How is Lightning v3.1 tested on the Humanness Index™?: Listeners hear Lightning v3.1 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.
Which languages does Lightning v3.1 support?: The model card lists 12 languages spanning English, Hindi, Spanish, and nine Indian languages, with automatic language identification and mid-sentence code-switching; the Lightning V3 launch post lists 15 including French, German, Italian, Dutch, Swedish, and Portuguese.

Keep exploring

Smallest.aiAll Smallest.ai models on the Index

Back to the Humanness Index™

How human does your model really sound?

The benchmark is open source. Suggest a model, read the methodology, or ask us to put your voice in the arena.

Add your model Star on GitHub

Lightning v3.1 key stats

Latency (measured)

420 ms¹

Languages

12²

Price / 1M chars

$15³

Streaming

Yes⁴

Voice cloning

instant, 5-15 s sample⁵

Released

January 2026⁶

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.smallest.ai/waves/model-cards/text-to-speech/lightning-v-3-1 (checked 2026-06-12) Model card: 12 languages with auto-detection and code-switching; the launch post lists 15 including major European languages.

smallest.ai/pricing (checked 2026-06-12) Lightning V3.1 pay-as-you-go at ~$0.0145 per 1k characters = $14.50 per 1M; the Lightning V3.1 Pro tier bills ~$0.0195 per 1k ($19.50 per 1M).

docs.smallest.ai/waves/model-cards/text-to-speech/lightning-v-3-1 (checked 2026-06-12) HTTP, SSE, and WebSocket surfaces on the unified Waves API.

docs.smallest.ai/waves/documentation/voice-cloning/instant-clone-api (checked 2026-06-12)

smallestai-ff1e543d.mintlify.app/v4.0.0/content/changelog/announcements (checked 2026-06-12, confidence: medium) Waves changelog lists the v3.1 release in January 2026; the Lightning V3 family launch post followed on 2026-03-27.

Background

At a glance

Rank

Provider

Model

Humanness

Latency

#12

ElevenLabs

Turbo v2

302 ms

#13

Neuphonic

neu_hq

276 ms

#14

Smallest.ai

Lightning v3.1

420 ms

#15

MiniMax

Speech 2 HD

357 ms

#16

ElevenLabs

Multilingual v2

1006 ms

Frequently asked questions

How is Lightning v3.1 tested on the Humanness Index™?

Listeners hear Lightning v3.1 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.

Which languages does Lightning v3.1 support?

The model card lists 12 languages spanning English, Hindi, Spanish, and nine Indian languages, with automatic language identification and mid-sentence code-switching; the Lightning V3 launch post lists 15 including French, German, Italian, Dutch, Swedish, and Portuguese.