Smallest AI

Real-time Voice AI. Built for Scale.

Smallest AISep 6, 2026

Real-time Voice AI. Built for Scale.

Smallest AI launch preview
212023633K227

Tags

<500KB2D (developers)ExplainerSeries AGlobalFeature launchVertical AI

TLVC Rating

2.20
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy

Community Rating

No ratings yet

to rate this launch.

Lightning v3.1 Pro is a text-to-speech model from Smallest AI aimed at teams building voice agents, IVR systems, and other real-time conversational products. The release focuses on the fine details that separate synthetic narration from a natural exchange, tuning where a pause lands and how a word is stressed so callers respond to the conversation rather than the model behind it. It is built for how voice systems actually run in production, generating speech in chunks , which matters for anyone deploying agents into live phone or support workflows. The Pro tier extends Lightning's coverage to 31 languages at 44.1 kHz with roughly 200ms time to first byte and streaming , and audio can be downsampled to 8 to 24 kHz for telephony . The stack also supports automatic language identification and code-switching mid-utterance across the trained voice set , which is useful for multilingual markets where a single call can shift between languages. Smallest AI was founded in 2023 by Sudarshan Kamath and Akshat Mandloi , with Kamath leading as CEO. The launch matters now because voice AI vendors are competing less on raw synthesis quality and more on conversational feel, and Lightning v3.1 Pro is the company's answer to that shift, positioned against providers like ElevenLabs, Cartesia, and OpenAI in the same category.

Comments (11)

Sign in to join the discussion.
P
Priya NadkarniSep 8, 2026

The pause placement thing is actually the whole game. Every TTS demo sounds fine until you hear it interrupt a customer mid-thought.

T
Tomás R.Sep 8, 2026

That jump cut at 0:14 in the demo video was rough, you can hear the room tone shift. Otherwise clean pacing on the voiceover.

A
Adaobi N.Sep 8, 2026

The tweet cuts off mid sentence which is either a cliffhanger or a scheduling tool bug. Either way I clicked, so nicely done.

K
Kwesi BoatengSep 8, 2026

Was the demo live or pre-rendered? Real-time claims deserve a stopwatch on screen.

R
Rania HaddadSep 8, 2026

Docs page or it didn't happen. Also please tell me streaming is a first-class SDK feature and not a footnote.

H
Hannelore M.Sep 8, 2026

Curious what the latency floor looks like under concurrent load. 'Built for scale' is doing a lot of heavy lifting in that tagline.

Y
yukiSep 8, 2026

Every voice AI launch: 'most human TTS ever built.' At some point one of you has to be lying.

F
Farhan I.Sep 8, 2026

Naive question but if the person doesn't care about the model, why is the model version in the product name?

M
Mira VolkovSep 8, 2026

Any word on data residency for EU deployments and whether voice inputs get retained for training? Compliance teams will ask before procurement will.

S
Sam O.Sep 8, 2026

What's the webhook story for call events and are rate limits per key or per org? Asking because I've been burned twice this month.

D
Deepthi V.Sep 8, 2026

Per-character pricing or per-second of audio? The margin math is very different and it matters a lot for anyone running high volume support flows.