Best AI for transcription and subtitles

There is no single best tool: ElevenLabs Scribe v2 stands out for language coverage and price, OpenAI's gpt-transcribe and Deepgram Nova-3 are strong API options, and Whisper large-v3 is the choice when audio must stay on your own computer. For videos already on YouTube, its automatic captions are a convenient starting point.

• ElevenLabs Scribe v2: more than 90 languages at 0.22 USD per hour via API. English, Spanish, French, German, Italian, Dutch, Portuguese, Turkish and Japanese sit in its 'Excellent' tier, a word error rate of 5 percent or lower (vendor claim).
• OpenAI: gpt-transcribe, the model OpenAI now recommends, costs 0.0045 USD per minute. Uploads are limited to 25 MB. whisper-1 and the gpt-4o transcribe models are deprecated, with shutdown on 26 February 2027.
• Deepgram Nova-3: 0.0043 USD per minute for pre-recorded monolingual audio, 0.0052 USD for multilingual.
• Google Cloud Speech-to-Text Chirp 3: generally available for en-US, es-ES, fr-FR, de-DE, it-IT, nl-NL, pt-BR, tr-TR and ja-JP, among others.
• YouTube automatic captions: 67 languages for long videos and Shorts, though YouTube notes quality may vary.
• Whisper large-v3: open source under Apache 2.0, 99 languages, runs locally so audio never leaves your machine.

For subtitles, make sure your workflow gives you timestamps and an SRT or VTT export. Whatever you choose, always proofread names, numbers and technical terms, because every system makes mistakes on them.