> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kymaapi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# ElevenLabs Flash v2.5

> ElevenLabs Flash v2.5 is a text-to-speech model optimized for ultra-low-latency audio generation. Developers building real-time voice agents or interactive conversational interfaces should use it when sub-100ms time-to-first-byte is required.

## Overview

This model converts text input into spoken audio across 32 languages. It operates on a per-character billing model and delivers audio with a time-to-first-byte of approximately 75ms, operating at half the per-character cost of the Multilingual v2 variant.

On Kyma, requests are routed through an OpenAI-compatible endpoint with automatic failover to maintain delivery. Prompt caching applies to repeated text prefixes, billing them at this model's cached input rate. Usage metrics return exact character costs in the response payload, and the X-Kyma-Model header confirms the executing model.

The model is strictly for speech synthesis and does not support reasoning, vision, or structured output formats. It is optimized for speed and cost efficiency rather than maximum vocal fidelity, and it processes text inputs up to a 5000-token context window.

## Specs

| Field              | Value                                                            |
| ------------------ | ---------------------------------------------------------------- |
| Model ID           | `eleven-flash-v2-5`                                              |
| Best for           | Real-time voice agents, conversational AI, low-latency narration |
| Context window     | 5,000                                                            |
| Input modalities   | Text                                                             |
| Output modalities  | Audio                                                            |
| Tool calling       | Yes                                                              |
| Structured outputs | Yes                                                              |
| Prompt caching     | Yes                                                              |
| Speed              | Fast                                                             |
| Cost band          | Cheap                                                            |
| Release stage      | Stable                                                           |

## Pricing

|        | Per 1M tokens |
| ------ | :-----------: |
| Input  |     \$0.20    |
| Output |     \$0.60    |

## Use this when

* **Real-Time Voice Agents** — Generates conversational audio with sub-100ms latency for interactive AI assistants.
* **Low-Latency Narration** — Streams spoken content quickly for dynamic reading applications.
* **Cost-Optimized TTS** — Reduces synthesis expenses for high-volume text-to-speech workloads.
* **Multilingual Conversational AI** — Supports voice generation across 32 languages for global user bases.

## Not ideal for

Do not use this model for applications requiring studio-grade vocal fidelity, complex emotional prosody, or non-speech audio generation.

## Pick something else when

* You need higher vocal quality and emotional range: use [`eleven-multilingual-v2`](/models/eleven-multilingual-v2) or [`eleven-v3`](/models/eleven-v3).
* You need to generate music or sound effects: use [`elevenlabs-music`](/models/elevenlabs-music) or [`elevenlabs-sfx`](/models/elevenlabs-sfx).
* You need to clone a specific voice or design custom vocal profiles: use [`minimax-voice-clone`](/models/minimax-voice-clone) or [`minimax-voice-design`](/models/minimax-voice-design).

## Example

```python theme={null}
from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="ky-...")

response = client.chat.completions.create(
    model="eleven-flash-v2-5",
    messages=[{"role": "user", "content": "..."}],
)
```

## Agent query example

Ask the API which models fit, instead of hardcoding an id:

```bash theme={null}
curl "https://kymaapi.com/v1/models?recommended_for=generate-speech&tools=true&quality_tier=strong"
```

## FAQ

**How does Kyma handle pricing for this model?**

Billing is calculated per character, and repeated prompt prefixes are charged at this model's cached input rate through automatic caching.

**Can I use this model for structured data output or reasoning tasks?**

No, it is strictly a text-to-speech model that outputs audio and does not support reasoning, vision, or JSON formatting.

**What happens if the serving path experiences high latency?**

Kyma automatically reroutes the request to a healthy serving path to maintain delivery, and the exact cost is always returned in the usage object.