# GPT transcription API: gpt-transcribe and gpt-4o-mini-transcribe speech-to-text

Source: https://ofox.io/docs/develop/models/gpt-transcribe

OpenAI's speech transcription models. Upload audio through `POST /v1/audio/transcriptions` and get text back.

**Model IDs (copy them exactly)**

- GPT Transcribe: `openai/gpt-transcribe` — Billed per second of audio
- GPT-4o mini Transcribe: `openai/gpt-4o-mini-transcribe` — Billed per token

For the common endpoint parameters, see [Audio Transcriptions API](https://ofox.io/docs/api/openai/audio-transcriptions). For live prices, see the [model page](https://ofox.io/models/openai/gpt-transcribe). If you need word-level timestamps or speaker diarization, use [ElevenLabs Scribe v2](https://ofox.io/docs/develop/models/elevenlabs#scribe).

## Examples

**Python**

```python filename="gpt_transcribe.py"
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.ofox.io/v1", api_key=os.environ["OFOX_API_KEY"])

with open("meeting.mp3", "rb") as f:
    result = client.audio.transcriptions.create(
        model="openai/gpt-transcribe",
        file=f,
        language="zh",
        prompt="以下是普通话的句子，请使用简体中文。",
    )
print(result.text)
print(result.usage)
```

**cURL**

```bash filename="Terminal"
curl https://api.ofox.io/v1/audio/transcriptions \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -F model=openai/gpt-transcribe \
  -F language=zh \
  -F prompt="以下是普通话的句子，请使用简体中文。" \
  -F file=@meeting.mp3
```

## Parameters and billing

**Key point**:When transcribing Mandarin, pass both language and a prompt that asks for Simplified Chinese; otherwise the output may contain Traditional characters.

| Item | Description |
|---|---|
| `language` and `prompt` | In our tests, with only `language: "zh"`, `gpt-transcribe` still output Traditional characters in 1 of 4 runs; after also passing `prompt: "以下是普通话的句子，请使用简体中文。"`, all 4 runs were in Simplified Chinese |
| `response_format` | Only `json` (default) is supported. `verbose_json`, `text`, `srt`, and `vtt` all return 400 |
| Response content | `text` and `usage`; no word-level timestamps |
| Billing | `gpt-transcribe` is billed per second of audio, with `usage` as `{"type": "duration", "seconds": ...}`; `gpt-4o-mini-transcribe` is billed per token, with `input_tokens` and `output_tokens` in `usage` |

In our tests, about 9 seconds of Chinese conversation was transcribed in about 2 to 3 seconds.

## Common errors
| HTTP | Error message | Cause | Fix |
|---|---|---|---|
| 400 | `response_format 'verbose_json' is not compatible with model 'gpt-transcribe'.` | GPT transcription models do not support `verbose_json` | Use `json` |
| 400 | `Unsupported response_format 'srt'. Only 'json' and 'verbose_json' are supported on this gateway` | `text`, `srt`, or `vtt` was used | Use `json` |
| 400 | `Unsupported file format txt` | The uploaded file is not audio | Upload audio such as `mp3` or `wav` |

For more errors, see [Audio Transcriptions API](https://ofox.io/docs/api/openai/audio-transcriptions#errors).

## Official docs
- [OpenAI · Speech to text](https://platform.openai.com/docs/guides/speech-to-text)

## FAQ

**GPT transcription outputs Traditional characters for Chinese?**

Pass both language: "zh" and prompt: "以下是普通话的句子，请使用简体中文。". In our tests, passing only language still occasionally produced Traditional characters; adding the prompt consistently produced Simplified Chinese.

**Can GPT transcription output srt subtitles?**

No, only json is supported. If you need timestamps to generate subtitles, use ElevenLabs Scribe v2, which returns start and end times for each word.

**How are gpt-transcribe and gpt-4o-mini-transcribe billed?**

gpt-transcribe is billed per second of audio; gpt-4o-mini-transcribe is billed per token, with usage shown in the usage field of the response. See the model page for rates.