Whisper Large v3 Turbo

Overview

This model converts audio input into text transcripts, operating at 228x real-time speed. It automatically detects the spoken language and returns transcripts with precise timestamps. The architecture is tuned for strong accuracy while maintaining a cheap cost tier. On Kyma, it runs through an OpenAI-compatible endpoint with automatic request failover. If a serving path degrades, traffic reroutes transparently. The model supports prompt caching, which bills repeated prefixes at this model’s cached input rate. Usage metadata, including exact request cost and the resolved model ID, is returned in the response headers and JSON payload. The model accepts audio input and outputs plain text. It does not support vision, structured outputs, or reasoning tasks. The context window caps at 3,600 tokens, making it suitable for standard-length audio clips rather than multi-hour continuous streams without chunking.

Specs

Field	Value
Model ID	`whisper-v3-turbo`
Best for	Transcription, captions, voice agents
Context window	3,600
Input modalities	Audio
Output modalities	Text
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

Live Meeting Transcription — Converts conference audio into timestamped text for searchable meeting notes.
Voice Agent Pipelines — Feeds real-time speech recognition into conversational AI workflows with minimal latency.
Automated Caption Generation — Produces synchronized subtitles for video content with automatic language detection.
Audio Content Indexing — Transforms podcast and interview recordings into searchable text archives.

Not ideal for

Do not use this model for multi-hour continuous audio streams without manual chunking, as the 3,600-token context window will truncate longer inputs.

Pick something else when

You need structured JSON output, or reasoning over the audio rather than a plain transcript: use gpt-4o-mini-transcribe-2025-12-15.

Example

from openai import OpenAI

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

response = client.chat.completions.create(
    model="whisper-v3-turbo",
    messages=[{"role": "user", "content": "..."}],
)

Agent query example

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

curl "https://kymaapi.com/v1/models?recommended_for=transcribe-audio&tools=true&quality_tier=strong"

FAQ

How does Kyma handle model routing for this endpoint? Every request uses automatic failover; if the primary serving path degrades, Kyma reroutes the request transparently and reports the exact model that ran in the X-Kyma-Model header. How is billing calculated for repeated audio prompts? Prompt caching applies to repeated prefixes, billing them at this model’s cached input rate. New accounts include a $0.50 signup credit to cover initial testing. Does this model support structured outputs or reasoning? No. It is strictly a speech-to-text model that outputs plain text transcripts with timestamps. It does not support structured outputs, vision, or reasoning tasks.

Alias	Resolves to
`transcribe`	`whisper-v3-turbo`

​Overview

​Specs

​Pricing

​Use this when

​Not ideal for

​Pick something else when

​Example

​Agent query example

​FAQ

​Related aliases