Speech-to-text model powered by GPT-4o mini.
GPT-4o Mini Transcribe is OpenAI's cost-efficient speech-to-text model, built on the GPT-4o Mini architecture. It transcribes audio input into text and offers improvements in word error rate (WER), language recognition, and overall accuracy compared to the original Whisper models. The model is designed for developers who need reliable transcription at a lower price point than the full-size GPT-4o Transcribe, while still benefiting from reinforcement-learning-enhanced accuracy.
OpenAI employs a reinforcement learning (RL) paradigm in training, pushing transcription accuracy to state-of-the-art levels while reducing hallucination. The model supports 99+ languages, handles accents, noisy environments, and varying speech speeds better than its Whisper predecessors. On the FLEURS multilingual benchmark, GPT-4o Transcribe (the full sibling) achieves single-digit WER across major languages; GPT-4o Mini Transcribe follows closely as a cost-optimised alternative.
| Type | Rate |
|---|---|
| Audio | $0.0032 / minute of audio |
| Input | $1.35 / 1M tokens |
| Output | $5.40 / 1M tokens |
| Audio input | $3.24 / 1M tokens |
| Estimated | $0.0032 / minute |
request is estimated price per minute
| Release date | March 20, 2025openai.com ↗ | verified |
| Supported languages | 99+ languages via ISO-639-1 language codesopenai.com ↗ | verified |
| Architecture | GPT-4o Mini with RL-enhanced pretraining for speechopenai.com ↗ | verified |
| Input audio formats | mp3, mp4, mpeg, mpga, m4a, wav, webm, flac, ogg, ogaplatform.openai.com ↗ | verified |
| Max file size | 25 MBplatform.openai.com ↗ | verified |
| Training data cutoff | Not publicly disclosed | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
FLEURS English WER FLEURS English WER for gpt-4o-transcribe (full-size sibling) is reported at 2.46% in the OpenAI blog post. Mini variant WER is expected to be higher but exact figures have not been independently published. | ~2.46% (gpt-4o-transcribe reference)% | openai.com ↗ | |
FLEURS Multilingual OpenAI reports single-digit WER across major languages on FLEURS. The gpt-4o-transcribe family match or outperform Whisper v2/v3 across all language evaluations. | Single-digit WER across major languages | openai.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/audio/transcriptions -H "Authorization: Bearer $THE_ROUTER_API_KEY" -F "model=openai/gpt-4o-mini-transcribe" -F "file=@speech.mp3"Transcribe audio files via the standard OpenAI /v1/audio/transcriptions endpoint through TheRouter. Supports multiple audio formats and optional language hints for improved accuracy.
curl https://api.therouter.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-F "file=@recording.mp3" \
-F "model=openai/gpt-4o-mini-transcribe" \
-F "response_format=json" \
-F "language=en"OpenAI announced gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-mini-tts — a new family of speech-to-text and text-to-speech models. The transcribe models employ RL-heavy training to achieve state-of-the-art WER, outperforming Whisper v2/v3 across FLEURS multilingual benchmarks.
GPT-4o Mini Transcribe uses GPT-4o Mini as its backbone with RL-enhanced pretraining, delivering significantly lower word error rates across all evaluated languages. It handles accents, noisy environments, and varying speech speeds better than Whisper. Whisper-1 remains available for legacy integrations, but the newer model is recommended for new projects.
No — this model only supports the standard /v1/audio/transcriptions endpoint, which processes uploaded audio files. For real-time streaming transcription, use OpenAI's Realtime API models (gpt-realtime-*) which support WebSocket-based audio streaming with lower latency.
Supported response formats include json (default), text, srt (SubRip subtitle format), verbose_json (with timestamps and segments), and vtt (WebVTT). The json and verbose_json formats offer structured output usable in downstream pipelines.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-28 | verified |
| Supported languages | openai.com ↗ | 2026-05-28 | verified |
| Architecture | openai.com ↗ | 2026-05-28 | verified |
| Input audio formats | platform.openai.com ↗ | 2026-05-28 | verified |
| Max file size | platform.openai.com ↗ | 2026-05-28 | verified |
| Training data cutoff | — | — | unknown |
| FLEURS English WER | openai.com ↗ | 2026-05-28 | to verify |
| FLEURS Multilingual | openai.com ↗ | 2026-05-28 | to verify |
| OpenAI launches next-generation audio models including GPT-4o Mini Transcribe | openai.com ↗ | 2026-05-28 | verified |
| What's the difference between GPT-4o Mini Transcribe and Whisper-1? | openai.com ↗ | 2026-05-28 | to verify |
| What is the output format of the transcription? | platform.openai.com ↗ | 2026-05-28 | to verify |