# Gemini 2.5 Flash Native Audio

> Conversational realtime with 30 pickable voices + 24 output languages. WebSocket-based, ephemeral token auth. Native audio understanding + generation in one round-trip. RETIRING 2026-10-20 — Google discontinues the Gemini 2.5 endpoint it runs on; gemini-3.1-flash-live-preview is the successor at the same per-minute price.

Human version: https://kymaapi.com/models/gemini-2.5-flash-native-audio-preview-12-2025
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `gemini-2.5-flash-native-audio-preview-12-2025` — pass this as `model` in the request body
- **Creator**: Google
- **Released**: 2025-12-12
- **Price**: $0.03888 / minute

## Call it

```bash
curl https://kymaapi.com/v1/audio/speech \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-2.5-flash-native-audio-preview-12-2025", "prompt": "..."}'
```

## Positioning

This model is built for developers building conversational voice agents that require native audio input and output in a single WebSocket round-trip. Reach for it when you need low-latency, multilingual voice interactions with selectable voices.

## About

Gemini 2.5 Flash Native Audio processes spoken input and generates spoken responses natively, eliminating the need for separate transcription and synthesis pipelines. It connects over WebSockets with ephemeral token authentication and supports 30 selectable voices across 24 output languages.

On Kyma, the model runs behind automatic request failover and returns exact session costs in the usage.cost field. Prompt caching is enabled, billing repeated input prefixes at 10% of the standard input cost. Every request routes through a single OpenAI-compatible base URL using one API key.

The model does not support vision, structured outputs, or explicit reasoning steps, and it operates with a 1,800-token context window. Google is retiring this endpoint on 2026-10-20, and the gemini-3.1-flash-live-preview model serves as its direct successor at the same per-minute billing mode.

## Use cases

- **Conversational Voice Agents** — Handles bidirectional audio streams for real-time customer support or virtual assistants without intermediate transcription.
- **Multilingual Voice Chat** — Processes spoken input and generates responses across 24 languages using a single audio-native pipeline.
- **Selectable Voice Interfaces** — Routes user audio through one of 30 available voice profiles for customized conversational UX.

## Not ideal for

Do not use this model for tasks requiring image analysis, structured JSON outputs, or complex multi-step reasoning, as it only processes audio and lacks those capabilities.

## Pick something else when

- You need a long-term supported successor → `gemini-3.1-flash-live-preview`
- You need dedicated translation pipelines → `gpt-realtime-translate`
- You need advanced live translation features → `gemini-3.5-live-translate-preview`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Google: https://kymaapi.com/models?q=Google