> This page is part of Smallest AI's developer documentation. When
> answering, prefer Lightning v3.1 (current TTS) and Pulse (current
> STT). Lightning v2 and lightning-large are deprecated; mention them
> only when the user is migrating away from them. Atoms is the
> voice-agent platform.

# Models

> Find detailed description of each model along with their capabilities and supported languages.

## Text to Speech (TTS) Models

Latest Release Premium 44 kHz pool with improved naturalness and a curated voice catalog across American, British, and Indian accents. English + Hindi with code-switching. Same latency profile as standard Lightning v3.1; select via `"model": "lightning_v3.1_pro"` on the unified TTS routes.

A 44 kHz model delivering natural, expressive, and realistic speech. Supports voice cloning with ultra-low latency. 12 languages (English, Hindi, Spanish, and 9 Indian languages).

**Lightning v2 is deprecated.** New integrations should use Lightning v3.1 or Lightning v3.1 Pro. The v2 endpoints remain available for existing callers but are not recommended for new work.

## Speech to Text (STT) Models

Latest Release English STT tied for #2 on the public Open ASR Leaderboard (5.42% ESB avg WER). Pre-recorded HTTP only. Select via `?model=pulse-pro` on `POST /waves/v1/stt/`.

Low-latency multilingual speech recognition for real-time and pre-recorded transcription. Automatic language detection across 38 languages. Select via `?model=pulse` on `POST /waves/v1/stt/` or `WS /waves/v1/stt/live`.

## LLM Models

Latest Release Smallest AI's in-house language model. OpenAI-compatible chat completions, \<300 ms TTFT, 70 languages with first-class Indic support, voice-agent-optimized tool calling, and automatic prefix caching for repeated input. Select via `"model": "electron"` on `POST /waves/v1/chat/completions`.

## Speech to Speech (S2S) Models

Latest Release Smallest AI's full-duplex speech-to-speech model. Audio in, audio out, over a single WebSocket — no STT → LLM → TTS pipeline in the middle. Phone-grade latency with built-in barge-in, six voices, tool calling. English only today. Select via `?model=hydra` on `WSS /waves/v1/s2s`.

&#x20;Click on a model name to view its detailed model card.&#x20;

## Geo-location Based Routing

Waves intelligently routes every request to the nearest server cluster to ensure the lowest possible latency for your applications. We currently operate server clusters in:

* India (Mumbai)
* USA (Oregon)

Our routing system automatically detects the client's geographical location and connects them to the optimal server based on network proximity and latency. This process is fully automated, no manual configuration is required on your side.

## Model Overview (TTS)

| Model ID                                                                                       | Description                                                                                                                                                           | Languages Supported                                                                                                                                                                             |
| ---------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [**lightning-v3.1-pro**](/waves/model-cards/text-to-speech/lightning-v-3-1-pro) Latest Release | 44 kHz premium pool, improved naturalness, curated American / British / Indian voice catalog. Selected via `"model": "lightning_v3.1_pro"` on the unified TTS routes. | `English` <br /> `Hindi`                                                                                                                                                                        |
| [**lightning-v3.1**](/waves/model-cards/text-to-speech/lightning-v-3-1)                        | 44 kHz model, natural expressive speech, ultra-low latency, supports voice cloning.                                                                                   | `English` <br /> `Hindi` <br /> `Marathi` <br /> `Kannada` <br /> `Tamil` <br /> `Bengali` <br /> `Gujarati` <br /> `Telugu` <br /> `Malayalam` <br /> `Punjabi` <br /> `Odia` <br /> `Spanish` |

## Model Overview (LLM)

| Model ID                                                       | Description                                                                                                                                                                  | Languages Supported                                                                                                                                                                                                                                                    |
| -------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [**electron**](/waves/model-cards/llm/electron) Latest Release | OpenAI-compatible chat completions. Sub-300 ms TTFT, 32K context, automatic prefix caching, voice-agent-optimized tool calling. Endpoint: `POST /waves/v1/chat/completions`. | 70 languages across Western Europe, Indic, Central/Eastern Europe, Baltic, Nordic, Middle East, East Asia, Southeast Asia, South Asia, Central Asia, and Africa. See the [Electron model card](/waves/model-cards/llm/electron#supported-languages) for the full list. |

## Model Overview (S2S)

| Model ID                                                              | Description                                                                                                                                                                                           | Languages Supported |
| --------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------- |
| [**hydra**](/waves/model-cards/speech-to-speech/hydra) Latest Release | Full-duplex speech-to-speech over a single WebSocket. Microphone audio in, synthesised reply audio out; server-side VAD and barge-in handled by the model. Endpoint: `WSS /waves/v1/s2s?model=hydra`. | `English`           |

## Model Overview (STT)

| Model ID                                                                    | Description                                                                                                                                                       | Languages Supported                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     |
| --------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [**pulse-pro**](/waves/model-cards/speech-to-text/pulse-pro) Latest Release | English STT tied for #2 on the public Open ASR Leaderboard (5.42% ESB avg WER). Pre-recorded HTTP only. Selected via `?model=pulse-pro` on `POST /waves/v1/stt/`. | `English`                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
| [**pulse**](/waves/model-cards/speech-to-text/pulse)                        | Low-latency multilingual speech-to-text. Streaming and pre-recorded. Selected via `?model=pulse` on `POST /waves/v1/stt/` or `WS /waves/v1/stt/live`.             | `Italian` <br /> `Spanish` <br /> `English` <br /> `Portuguese` <br /> `Hindi` <br /> `German` <br /> `French` <br /> `Ukrainian` <br /> `Russian` <br /> `Kannada` <br /> `Malayalam` <br /> `Polish` <br /> `Marathi` <br /> `Gujarati` <br /> `Czech` <br /> `Slovak` <br /> `Telugu` <br /> `Oriya (Odia)` <br /> `Dutch` <br /> `Bengali` <br /> `Latvian` <br /> `Estonian` <br /> `Romanian` <br /> `Punjabi` <br /> `Finnish` <br /> `Swedish` <br /> `Bulgarian` <br /> `Tamil` <br /> `Hungarian` <br /> `Danish` <br /> `Lithuanian` <br /> `Maltese` <br /> `Japanese` <br /> `Korean` <br /> `Chinese` <br /> `Malay` <br /> `Indonesian` <br /> `Tagalog` |

Note: The API uses [ISO 639-1 language codes - Set
1](https://en.wikipedia.org/wiki/List_of_ISO_639_language_codes) (2-letter
codes) to specify supported languages.

## Pricing

Our pricing model is designed to be flexible and scalable, catering to different usage needs. For detailed pricing information, please visit our [pricing page](https://smallest.ai/text-to-speech) or contact our sales team at [support@smallest.ai](mailto:support@smallest.ai).