# Whisper

Whisper made large-scale speech recognition an API call. Trained on diverse multilingual audio, Whisper transcribes, translates to English, and identifies languages with one model.

- **Model ID:** `openai/whisper-1`
- **Type:** transcription
- **Providers:** openai
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** $0.0001/1M input tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/whisper-1

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { experimental_transcribe as transcribe } from 'ai';
import { gateway } from '@ai-sdk/gateway';
import { readFile } from 'node:fs/promises';

const result = await transcribe({
  model: gateway.transcriptionModel('openai/whisper-1'),
  audio: await readFile('audio.mp3'),
});
```

## About

OpenAI released Whisper as an open-source speech recognition system in September 2022, trained on 680,000 hours of multilingual and multitask supervised data collected from the web. That scale and diversity made Whisper robust to accents, background noise, and technical language. The hosted API model, Whisper, arrived on September 21, 2022 alongside the GPT-3.5 Turbo API, serving the large-v2 weights through a managed endpoint.

Whisper is a multitask model. Beyond multilingual transcription, Whisper performs speech translation into English and language identification, so one model covers the common speech workflows without separate systems.

Because the underlying model is open source, Whisper became the reference point for speech-to-text across the industry. Teams that run open-source Whisper locally can use Whisper through AI Gateway for the hosted path, keeping output behavior familiar while gaining unified authentication, observability, and spend controls.

## What to consider

Newer transcription models exist in the catalog: `gpt-4o-transcribe` and `gpt-4o-mini-transcribe` improve word error rate over Whisper-generation models. Whisper remains the pick when you want the multitask capabilities, duration-based pricing, or consistency with self-hosted open-source Whisper deployments.

## When to use

### Best For

- Multilingual transcription where one model must handle many languages without configuration
- Speech translation workflows that turn non-English audio into English text
- Language identification as a preprocessing step in audio pipelines
- Duration-priced transcription where billing by audio length simplifies cost modeling
- Matching behavior with self-hosted open-source Whisper deployments

### Consider Alternatives When

- Transcript accuracy is the priority; `gpt-4o-transcribe` and `gpt-4o-mini-transcribe` improve word error rate over Whisper-generation models
- You're building live voice experiences; the gpt-realtime family handles speech-to-speech conversation

## Best for

- **Multilingual transcription:** One model handles many languages without per-language configuration
- **Speech translation:** Non-English audio translated directly into English text
- **Language identification:** Detecting the spoken language as a pipeline preprocessing step
- **Duration-based billing:** Costs scale with audio length, which simplifies budget modeling
- **Open-source consistency:** Hosted behavior that matches self-run Whisper deployments

## Consider alternatives

- **Accuracy-critical transcripts:** `gpt-4o-transcribe` and `gpt-4o-mini-transcribe` deliver lower word error rates
- **Live voice experiences:** The gpt-realtime family handles speech-to-speech conversation directly
- **High-volume token pricing:** The newer transcribe models bill by tokens if that fits your cost model better

## Frequently asked questions

### How do I access Whisper through AI Gateway?

Use the AI SDK's `transcribe` function from the `ai` package with this model's slug, passing audio as a file buffer, base64 string, or URL. Audio support on AI Gateway is in beta. Authenticate with an AI Gateway API key or OIDC token.

### What tasks can Whisper perform beyond transcription?

Whisper is a multitask model: multilingual speech recognition, speech translation into English, and language identification. That breadth is what distinguishes Whisper from single-purpose transcription models.

### Is Whisper the same as open-source Whisper?

Whisper serves OpenAI's large-v2 Whisper weights through a hosted endpoint. Teams running open-source Whisper can expect familiar transcription behavior from the managed model.

### How does Whisper compare to gpt-4o-transcribe?

GPT-4o Transcribe and GPT-4o mini Transcribe improve word error rate and language recognition over the original Whisper models. Whisper counters with multitask coverage, translation and language identification, and duration-based pricing.

### How is Whisper priced?

Transcription is billed by audio duration. Current rates are listed on this page and update when providers change list prices.

### Does Whisper support zero data retention through AI Gateway?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

### What are typical latency characteristics?

This page shows live performance metrics measured across real AI Gateway traffic.

## Links

- [Model page](https://vercel.com/ai-gateway/models/whisper-1)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://developers.openai.com/api/docs/models/whisper-1)