# Smallest AI review (2026): What it is and why enterprises choose Bland Speech

By Samuel Oram-Jones

Smallest AI retires its Lightning v3.1 text to speech model on 30 November 2026 at 18:30 UTC, and Smallest AI says voice clones made on that model are affected by the retirement. Smallest AI tells teams on the model to move to Lightning v3.1 Pro before then.

Lightning v3.1 Pro costs about $19.50 per million characters, less than half of Bland Speech's $40, and Smallest AI states its latency two different ways. As of November 2026, that leaves a buyer weighing a cheaper and faster-sounding API against a model line that changes names and versions quickly.

## What is Smallest AI and how does it work?

Smallest AI sells a realtime voice AI suite for enterprise: voice agents, Lightning text to speech, Pulse speech to text, Hydra, a full-duplex speech to speech model in beta, and the Electron language model. Lightning is the product relevant to a team shopping for a voice, and it is reached through an API.

### Which models are current, and which are retiring?

Lightning v3.1 Pro is the current model, which Smallest AI describes as premium curated voices with Hindi code-switching across 31 languages. Lightning v3.1, which has 217 voices, retires on 30 November 2026. Lightning v2 and lightning-large are already deprecated, and a preview checkpoint, `lightning_v3.1_pro_10_26`, covers English, Spanish, and Hindi and becomes stable on 28 November 2026. Smallest AI's July 2026 changelog says Pro went multilingual with 27 new languages and 149 new voices.

A team that built on a retiring model has a short migration window, and any clone made on Lightning v3.1 is part of it.

### How does Lightning stream, and how fast is it?

Smallest AI streams over HTTP, server-sent events, and a WebSocket. On the WebSocket, text can arrive in fragments that share a `context_id`, and Smallest AI buffers them, joins them at natural sentence boundaries, and speaks them as one continuous generation. Per-word timing events arrive interleaved with the audio.

Latency is where the numbers part ways. Smallest AI markets sub 100ms latency and "Sub 100ms time-to-first-audio," and says the first chunk arrives in about 100 ms. Smallest AI's latency benchmark reports about 200 ms to the first byte for both v3.1 and Pro at 40 concurrent requests over a WebSocket, measured from a client in the same AWS region as the inference pool, with the client round trip added. Bland publishes 199ms to first audio. Each figure is the company's own, so none of them is a race, and Smallest AI's benchmark figure and Bland's are close while its marketing figure is lower.

### What phone audio does it return?

Smallest AI returns PCM, WAV, MP3, or mu-law, at sample rates from 8 kHz up to a native 44.1 kHz. Mu-law is the 8-bit encoding phone networks carry. Smallest AI's agent platform also offers SIP trunking and Smallest-managed phone numbers.

### How many languages does it cover?

The count depends on where Smallest AI states it: 31 languages across two voice pools for the text to speech API, with Pro covering all 31, 29 for the Pro pool in a July 2026 changelog entry, 70 or more on its home page, and 15 in its text to speech marketing. A buyer needing a specific language should test that language on the model, not rely on a count.

## How much does Smallest AI cost?

Smallest AI's model pricing is about $0.175 per 10,000 characters for Lightning v3.1 and about $0.195 for Lightning v3.1 Pro, which works out to $17.50 and $19.50 per million characters. Pay as you go starts with $10 in free credits and no commitment, with a cap of 15 concurrent TTS streams, 100 requests a minute, and 450 million characters a month.

Enterprise pricing is custom and adds a 99.99% uptime SLA, on-premise deployment, SOC 2 compliance, and professional voice clone support. Pay as you go carries no uptime SLA. A HIPAA zero data retention option costs $1,000 a month as an add-on on pay as you go and is included on Enterprise.

Bland Speech is $0.04 per 1,000 characters, which is $40 per million, pay as you go with no monthly plan and 50,000 free characters once for a new account. On rate alone Smallest AI is the cheaper API, by more than half for Pro.

## Which sounds more human, Smallest AI or Bland Speech?

Smallest AI's Lightning v3.1 Pro ranked 11th of 13 AI voices at 952 in Design Arena's Audio Realism Bench results of 4 August 2026, while Bland Speech v3 ranked first at 1365 and the human recording scored 1500. In [Design Arena's published results](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/), vetted native speakers heard two recordings of one script and chose the voice that sounded more like a real person. Smallest AI's preview checkpoint, `lightning_v3.1_pro_10_26`, wasn't among those 13 voices.

The controls differ as well. The ones Smallest AI publishes are a speed multiplier and pronunciation dictionaries for brand names, technical terms, and acronyms. Bland Speech takes performance tags such as [laughs], [sighs], and [say warmly], two numeric controls for expressiveness and stability, and timed pauses from 0.1 to 10 seconds. Smallest AI reports listener ratings for expressiveness against other vendors from its own benchmark, which is separate from Design Arena's test.

For cloning, Smallest AI makes an instant clone from 5 to 15 seconds of clean audio, under 5 MB, through the API, the dashboard, or a Python SDK, and offers professional cloning on Enterprise. Bland Speech needs one clean sample of about 10 seconds for an instant clone, fine-tunes a professional clone on 30 minutes or more of audio after a consent and rights review, and keeps every clone private to the customer's organization.

## Is Smallest AI secure enough for enterprise calls?

Smallest AI places its security features on Enterprise. Smallest AI includes SOC 2 in its Enterprise compliance features with SSO and RBAC, zero data retention is available on Enterprise with a retention window of 24 hours to 90 days or none at all for the text to speech, speech to text, and language model APIs, and a self-host guide covers running Lightning and Pulse on a team's own servers with Docker or Kubernetes. Smallest AI offers on-premise deployment on Enterprise and states a 99.99% uptime SLA there.

On the Bland side, the company is SOC 2 Type II audited and HIPAA compliant with a signed BAA, and it was granted FedRAMP 20x Class C at Moderate on 17 November 2026. Bland states that speech synthesis runs inside that FedRAMP boundary. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. Enterprise pricing is custom, set with Bland's sales team.

## Is Smallest AI worth it for enterprises?

For an enterprise phone line, Bland Speech is the better buy: Bland Speech v3 scored 1365 in the blind tests Design Arena published on 4 August 2026, 413 points above Lightning v3.1 Pro at 952.

| | Bland Speech | Smallest AI |
|:--|:--|:--|
| Price | $0.04 per 1,000 characters ($40 per million) | About $19.50 per million for Pro, $17.50 for v3.1 |
| Free to start | 50,000 characters once | $10 in free credits |
| Latency, as published | 199ms to first audio | Sub 100ms in marketing, about 200 ms in a benchmark |
| Streaming | HTTP chunks and a WebSocket | HTTP, SSE, and a WebSocket |
| Phone-ready output | Mu-law at 8 kHz, PCM at 8 to 48 kHz | Mu-law, PCM, WAV, and MP3 from 8 to 44.1 kHz |
| Realism result of 4 August 2026 | First of 13 at 1365 | 11th of 13 at 952 |
| Instant clone | One clean sample of about 10 seconds | 5 to 15 seconds of clean audio |

<details><summary>What these rows mean</summary>

- **Price:** Smallest AI states prices per 10,000 characters, and the per-million figures are that rate multiplied by 100.
- **Latency:** Time to first audio is the gap between a request and the first sound. The two columns use each company's own method, and Smallest AI's benchmark adds client round trip.
- **Realism result:** Scores are Design Arena's published results of 4 August 2026, ranked among 13 AI voices.
- **Instant clone:** An instant clone is built from a short sample. Bland says every clone is private to a customer's organization and that the customer confirms the voice is theirs to use.

</details>

### Smallest AI pros and cons

<div class="cmp-proscons" data-rival="smallest-ai"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>About $19.50 per million characters for Lightning v3.1 Pro</li><li>HTTP, SSE, and WebSocket streaming, with per-word timing events</li><li>Mu-law at 8 kHz for phone audio</li><li>31 languages stated for the API, with Hindi code-switching</li><li>$10 in free credits and pronunciation dictionaries</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Lightning v3.1 retires on 30 November 2026, and clones made on it are affected</li><li>Ranked 11th of 13 AI voices at 952 in Design Arena's results of 4 August 2026</li><li>Latency is stated as sub 100ms in marketing and about 200 ms in a benchmark</li><li>Pay as you go carries no uptime SLA</li><li>The language count is stated as 14, 15, 29, 31, and 70 or more in different places</li><li>Professional cloning sits on Enterprise</li></ul></div></div>

### Why Bland Speech is the Smallest AI alternative for phone calls

Bland Speech gives up some of Smallest AI's price advantage and gets a dated blind result in return: v3 ranked first of 13 AI voices in Design Arena's results of 4 August 2026, the API picks the model from the voice a team passes, and the older `btts-2` still synthesizes. Bland Speech returns mu-law at 8 kHz, streams over HTTP chunks and a WebSocket that takes text as a language model writes it, and lets a new context interrupt a turn that is playing, which is how barge-in works.

Pipecat has a first-party Bland service, LiveKit Agents has a Python plugin, and an OpenAI-compatible endpoint lets a team switch a client by changing a base URL, a key, and a model.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>First of 13 AI voices in Design Arena's results of 4 August 2026, at 1365</li><li>199ms to first audio, as Bland publishes it</li><li>Mu-law at 8 kHz and a WebSocket with barge-in</li><li>One standard rate per character, with no monthly plan</li><li>First-party Pipecat service, a LiveKit Agents plugin, and an OpenAI-compatible endpoint</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a prototype needs, because Bland Speech is built for production calls</li></ul></div></div>

### Verdict: Smallest AI or Bland Speech for enterprise calls?

Bland Speech is the choice for phone calls and voice agents, because a buyer pays about twice Smallest AI's per-character rate for a voice that placed ten places higher in the same blind test. Smallest AI suits a team for which cost per character decides everything. A [Bland Speech vs Smallest AI comparison](/compare/bland-speech-vs-smallest-ai) sets the two side by side, [Smallest AI alternatives](/blog/smallest-ai-alternatives) covers the wider field, and [Cartesia vs Smallest AI](/blog/cartesia-vs-smallest-ai) sets Smallest AI against Cartesia. Bland Speech's [text to speech API](/text-to-speech-api) has the endpoints, the [text to speech comparison hub](/compare/text-to-speech) and [review index](/compare/text-to-speech/reviews) list more reviews, and a team can [try Bland Speech](/speech) with 50,000 free characters.

## Frequently asked questions

### How much does Smallest AI cost?

Smallest AI prices Lightning v3.1 Pro at about $0.195 per 10,000 characters, which is $19.50 per million, and Lightning v3.1 at about $0.175 per 10,000, or $17.50 per million. Pay as you go starts with $10 in free credits. Bland Speech is $0.04 per 1,000 characters.

### What is Lightning v3.1 Pro?

Lightning v3.1 Pro is Smallest AI's current text to speech model, with premium curated voices, Hindi code-switching, and 31 languages. It replaces Lightning v3.1, which retires on 30 November 2026 at 18:30 UTC.

### How fast is Smallest AI?

Smallest AI says sub 100ms to first audio in its marketing and reports about 200 ms to the first byte at 40 concurrent requests in its own benchmark. Bland publishes 199ms to first audio, and each number is the company's own, measured its own way.

### Does Smallest AI support phone audio?

Smallest AI returns mu-law, PCM, WAV, and MP3 from 8 kHz up to 44.1 kHz, and its agent platform offers SIP trunking and managed phone numbers. Bland Speech returns mu-law at 8 kHz and PCM from 8 to 48 kHz.

### Is Smallest AI SOC 2 compliant?

Smallest AI includes SOC 2 in the compliance features of its Enterprise plan, alongside SSO and RBAC. Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA.

### What is the best Smallest AI alternative?

Bland Speech is the alternative for phone calls and voice agents, with a first-place blind result dated 4 August 2026, a published 199ms to first audio, and mu-law output. A team ranking options on speed can read best low-latency text to speech APIs.
