# GPT-4o Transcribe

GPT-4o Transcribe is the accuracy-focused option in OpenAI's transcription lineup. Error rates drop compared to Whisper-generation models, especially on accents, background noise, and fast or slow speech.

- **Model ID:** `openai/gpt-4o-transcribe`
- **Type:** transcription
- **Providers:** openai
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** $2.5/1M input tokens, $10/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gpt-4o-transcribe

## 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/gpt-4o-transcribe'),
  audio: await readFile('audio.mp3'),
});
```

## About

GPT-4o Transcribe launched on March 13, 2024 as part of OpenAI's next-generation audio models for the API. The model builds on the GPT-4o architecture and was extensively pretrained on specialized audio-centric datasets.

GPT-4o Transcribe improves word error rate over the Whisper model family across established benchmarks. On the FLEURS multilingual benchmark, GPT-4o Transcribe outperforms Whisper v2 and Whisper v3 across language evaluations. OpenAI attributes the advances to targeted reinforcement learning and midtraining with diverse, high-quality audio.

The practical effect is fewer misrecognitions in the conditions that break transcription products: heavy accents, noisy rooms, and varying speech speeds. Through AI Gateway, GPT-4o Transcribe runs behind the same key, observability, and spend controls as the rest of your model traffic.

## What to consider

Audio support on AI Gateway is in beta. GPT-4o Transcribe runs through the AI SDK's `transcribe` function, which accepts a file buffer, base64 string, or URL. For high-volume pipelines where cost per request dominates, benchmark `gpt-4o-mini-transcribe` first; the mini variant often gets close on clean audio.

## When to use

### Best For

- Production transcription where word error rate directly affects product quality
- Recordings with heavy accents, background noise, or variable speech speeds
- Support, sales, and compliance transcripts where misrecognitions carry real cost
- Multilingual audio, backed by strong FLEURS results across language evaluations

### Consider Alternatives When

- Cost per request dominates and `gpt-4o-mini-transcribe` is accurate enough for your recordings
- You need speech translation to English or language identification; `whisper-1` handles both
- You're building live voice agents; the gpt-realtime family covers speech-to-speech

## Best for

- **Production transcription quality:** Word error rate directly affects the product and cleanup budgets
- **Challenging audio conditions:** Heavy accents, background noise, and variable speech speeds
- **High-stakes transcripts:** Support, sales, and compliance recordings where misrecognitions carry real cost
- **Multilingual workloads:** Strong FLEURS benchmark results across language evaluations

## Consider alternatives

- **Cost-driven pipelines:** `gpt-4o-mini-transcribe` handles clean audio well at a lower rate
- **Speech translation needs:** `whisper-1` covers translation to English and language identification
- **Live voice agents:** The gpt-realtime family handles speech-to-speech conversation directly

## Frequently asked questions

### How do I access GPT-4o Transcribe through AI Gateway?

Use the AI SDK's `transcribe` function from the `ai` package with this model's slug. Audio support on AI Gateway is in beta. Authenticate with an AI Gateway API key or OIDC token; no OpenAI credentials are needed in your application.

### How does GPT-4o Transcribe compare to Whisper?

GPT-4o Transcribe improves word error rate over the original Whisper models and outperforms Whisper v2 and Whisper v3 on the FLEURS multilingual benchmark. The gains are largest on accents, noisy environments, and varying speech speeds.

### When should I use gpt-4o-mini-transcribe instead?

Use the mini variant when volume and cost dominate and your audio is relatively clean. Reserve GPT-4o Transcribe for recordings where accuracy failures are expensive: noisy calls, strong accents, or compliance-grade transcripts.

### Can I stream transcription results through AI Gateway?

AI Gateway transcription is batch today. Send a complete file buffer, base64 string, or URL with the `transcribe` function and receive the transcript in the response.

### How is GPT-4o Transcribe priced?

Pricing is listed on this page. Rates reflect the providers routing through AI Gateway and update when providers change list prices.

### Does GPT-4o Transcribe 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/gpt-4o-transcribe)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://developers.openai.com/api/docs/models/gpt-4o-transcribe)