# Gladia — Full LLM Context

> Extended documentation for high-context LLMs. See also /llms.txt for the complete site map.

## Solaria-3 ASR Model

- **URL**: https://www.gladia.io/solaria-3
- **Type**: Production-grade automatic speech recognition (ASR) model
- **Provider**: Gladia (https://www.gladia.io)

### Positioning

Solaria-3 is built for production audio — noisy, fast-paced, and conversational. It targets business scenarios where accuracy on real customer recordings matters more than breadth of language coverage. Solaria-1 remains the choice for 100+ languages, code-switching, formal institutional speech, and real-time streaming.

### Supported languages (optimised)

English (EN), French (FR), German (DE), Spanish (ES), Italian (IT). Launch languages: EN and FR.

### Key benchmark results (WER — lower is better)

| Dataset | Solaria-3 WER | Rank | vs Solaria-1 |
|---|---|---|---|
| Real customer audio — English | 9.6% | #1 | −26% |
| Earnings22 cleaned AA | 6.4% | #1 | −21% |
| Switchboard | 33.9% | #1 | −9% |
| Noisy audio | 1.4% | #2 | −26% |
| Common Voice 24 | 6.9% | — | −16% |
| FLEURS | 3.7% | — | −5% |

### Multi-provider comparison (WER % — lower is better)

Full tables are embedded in the page HTML at https://www.gladia.io/solaria-3.

**Real customer audio — English**

| Provider | WER (%) |
|---|---|
| Solaria-3 | 9.6 |
| ElevenLabs Scribe v2 | 9.9 |
| AssemblyAI | 10.0 |
| Deepgram Nova-3 | 10.7 |
| Mistral Voxtral | 12.2 |
| Solaria-1 | 12.9 |

**Earnings22 Cleaned AA**

| Provider | WER (%) |
|---|---|
| Solaria-3 | 6.4 |
| AssemblyAI | 6.9 |
| ElevenLabs Scribe v2 | 7.7 |
| Speechmatics | 7.8 |
| Mistral Voxtral | 7.9 |
| Solaria-1 | 8.1 |
| Deepgram Nova-3 | 12.0 |

**Switchboard**

| Provider | WER (%) |
|---|---|
| Solaria-3 | 33.9 |
| Solaria-1 | 37.3 |
| AssemblyAI | 42.3 |
| Speechmatics | 46.0 |
| Mistral Voxtral | 48.1 |
| Deepgram Nova-3 | 49.8 |
| ElevenLabs Scribe v2 | 55.2 |

**Noisy audio**

| Provider | WER (%) |
|---|---|
| Mistral Voxtral | 1.0 |
| Solaria-3 | 1.4 |
| Solaria-1 | 1.9 |
| Speechmatics | 1.9 |
| AssemblyAI | 2.1 |
| Deepgram Nova-3 | 3.2 |
| ElevenLabs Scribe v2 | 4.0 |

### European language improvement (Solaria-3 vs Solaria-1, negative = better)

| Language | Real customer audio | Common Voice 24 |
|---|---|---|
| English (EN) | −26% | −16% |
| French (FR) | −18% | −19% |
| Italian (IT) | −10% | −12% |
| Spanish (ES) | −9% | ≈ flat |
| German (DE) | −3% | −13% |

### Known regressions vs Solaria-1 (published openly)

- VoxPopuli Cleaned AA: 2.9% (+32% vs Solaria-1 at 2.2%)
- Multilingual LibriSpeech: 8.0% (+36% vs Solaria-1 at 5.9%)

### Best use cases

- Business audio and call center recordings
- Noisy production environments (compression artefacts, variable mic quality)
- European multilingual deployments (EN, FR, DE, ES, IT)
- Entity-heavy transcription (company names, numbers, proper nouns)

### When to choose Solaria-1 instead

- 100+ language coverage or code-switching
- Formal parliamentary / institutional speech (VoxPopuli)
- Clean read-speech (Multilingual LibriSpeech)
- Real-time streaming with <103ms partials (Solaria-3 is async only for now)

### Availability & trial

- Status: Free trial period → GA
- Promo code: TRY-SOLARIA-3 — $200 worth of async transcription hours, once per account, redeem at https://app.gladia.io (must be used by June 21st 2026)
- On-premise deployment: Available

### API access

Select Solaria-3 as the model in the Gladia app or API for async transcription.