Best AI model to transcribe audio
Turning meetings, interviews and videos into text, or making subtitles. There is no open transcription quality leaderboard in the sources checked, so the list is ordered by price, not quality, with an open weights alternative.
Suggested models
Note: this list is ordered by price, not quality. There is no open quality leaderboard in the sources checked when you want to transcribe audio, so it takes one main model per company, from the lowest to the highest price per minute of audio on each company's official page.
1 Grok Speech to Text xAI
Why it fits: xAI's main model to transcribe audio. According to the official page: Languages: over 38, including Brazilian Portuguese · Speaker separation: yes · Real time: yes, over WebSocket · Also: word timestamps and files up to 500 MB (official page)
Ordered by price, not by quality.
- Price: US$ 0.10 per hour (that's US$ 0.001667 per minute) (source: xAI, official page; file upload; real time is US$ 0.20 per hour)?what it is
- Estimated cost (see the How we calculate block): about US$ 0.10 (estimate)
- Weights: Undisclosed?
- Alternative: open weights: Whisper large-v3
2 ElevenLabs Scribe v3 ElevenLabs
Why it fits: ElevenLabs's main model to transcribe audio. According to the official page: Languages: over 90 · Speaker separation: yes, up to 32 speakers · Real time: in the Scribe v3 Live version · Also: up to 1,000 key terms (official page)
Ordered by price, not by quality.
- Price: US$ 0.20 per hour (that's US$ 0.003333 per minute) (source: ElevenLabs, official page)?
- Estimated cost (see the How we calculate block): about US$ 0.20 (estimate)
- Weights: Undisclosed?
- Alternative: open weights: Whisper large-v3 · cheaper: Grok Speech to Text, US$ 0.001667 per minute
3 AssemblyAI Universal-3.5 Pro AssemblyAI
Why it fits: AssemblyAI's main model to transcribe audio. According to the official page: Languages: 18, including Portuguese and regional accents · Speaker separation: not stated · Real time: not stated · Also: accepts prompts and key terms (official page)
Ordered by price, not by quality.
- Price: US$ 0.21 per hour (that's US$ 0.0035 per minute) (source: AssemblyAI, official page)?
- Estimated cost (see the How we calculate block): about US$ 0.21 (estimate)
- Weights: Undisclosed?
- Alternative: open weights: Whisper large-v3 · cheaper: Grok Speech to Text, US$ 0.001667 per minute
4 gpt-transcribe OpenAI
Why it fits: OpenAI's main model to transcribe audio. According to the official page: Languages: detects the language and accepts language hints · Speaker separation: not stated · Real time: not stated · Also: files up to 25 MB, with prompts and keywords (official page)
Ordered by price, not by quality.
- Price: US$ 0.0045 per minute (source: OpenAI, official page)?
- Estimated cost (see the How we calculate block): about US$ 0.27 (estimate)
- Weights: Undisclosed?
- Alternative: open weights: Whisper large-v3 · cheaper: Grok Speech to Text, US$ 0.001667 per minute
5 MiniMax ASR MiniMax
Why it fits: MiniMax's main model to transcribe audio. According to the official page: Languages: multilingual, with automatic detection · Speaker separation: yes · Real time: yes, by streaming · Also: exports SRT and VTT subtitles (official page)
Ordered by price, not by quality.
- Price: US$ 0.38 per hour (that's US$ 0.006333 per minute) (source: MiniMax, official page)?
- Estimated cost (see the How we calculate block): about US$ 0.38 (estimate)
- Weights: Undisclosed?
- Alternative: open weights: Whisper large-v3 · cheaper: Grok Speech to Text, US$ 0.001667 per minute
How we calculate
Change the numbers and the cost of every model on the page is recalculated right away. Without JavaScript, the default values below apply.
With these numbers: 60 minutes of audio (estimate).
Cost of each model = quantity × the price per minute of audio published by the company. When the company uses another unit, the conversion is shown next to the price. It is a reference calculation and does not change the order of the list.
Other paid models we checked
Other versions and companies, with the price from each company's official page. They are left out of the list above when another model from the same company is already among the first five, when they do not meet this page's criterion, or when the company does not publish a price per minute of audio.
| Model | Price and source | Estimated cost |
|---|---|---|
| gpt-4o-mini-transcribe OpenAI | US$ 0.003 per minute (source: OpenAI, official page) | about US$ 0.18 (estimate) |
| Whisper (API) OpenAI | US$ 0.006 per minute (source: OpenAI, official page) | about US$ 0.36 (estimate) |
| Deepgram Nova-3 Deepgram | US$ 0.0077 per minute (source: Deepgram, official page; monolingual, pay as you go; the price shown is for streaming) | about US$ 0.46 (estimate) |
FAQ
What is the best AI model to transcribe audio?
By this page's rule, the list starts with Grok Speech to Text, ElevenLabs Scribe v3, AssemblyAI Universal-3.5 Pro, gpt-transcribe, MiniMax ASR. Note: this list is ordered by price, not quality. There is no open quality leaderboard in the sources checked when you want to transcribe audio, so it takes one main model per company, from the lowest to the highest price per minute of audio on each company's official page.
How much does it cost to transcribe audio with AI?
With the default values in the How we calculate block, the estimate ranges from US$ 0.10 to US$ 0.38 among the listed models with a published price. It is only an estimate and you can adjust the numbers on the page.
Is there an open weights option to transcribe audio?
Yes: Whisper large-v3.
Does this page use AI to recommend?
No. The recommendation comes from fixed rules over public data from Epoch AI, OpenRouter, Hugging Face and each company's official pricing page, in the latest version we checked. The typed text is only compared with keywords.
Terms used on this page
Use cases with their own page
Official pricing pages checked
- ElevenLabs, latest version we checked.
- AssemblyAI, latest version we checked.
- OpenAI, latest version we checked.
- MiniMax, latest version we checked.
- Deepgram, latest version we checked.
- xAI, latest version we checked.
Sources
- Hugging Face Hub, metadados públicos do Hub; licença de cada modelo no repositório, latest version we checked.
Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.