Audio token estimates
Audio Token Calculator
Convert audio duration into tokens and transcription cost. Enter a length or drop in an audio file — the file never leaves your browser.
Audio length
Common lengths
Duration
10m 0s
600 seconds
Cheapest transcription
$0.0058
on Gemini 3.5 Flash-Lite (audio input)
Most audio tokens
24,000
on gpt-4o-transcribe
| Model | Billing | Audio tokens | Est. cost |
|---|---|---|---|
gpt-transcribe OpenAI | $0.0045/min | — | $0.045 |
gpt-live-transcribe OpenAI | $0.017/min | — | $0.170 |
gpt-live-1 OpenAI | $0.050/min | — | $0.500 |
gpt-realtime-translate OpenAI | $0.034/min | — | $0.340 |
gpt-4o-transcribeDeprecated OpenAI | $0.006/min | 24,000 | $0.060 |
gpt-4o-mini-transcribeDeprecated OpenAI | $0.003/min | 24,000 | $0.030 |
whisper-1Deprecated OpenAI | $0.006/min | — | $0.060 |
gpt-4o-transcribe-diarizeDeprecated OpenAI | $0.006/min | 24,000 | $0.060 |
Gemini 3.8 Live (audio input) | $3/1M audio tokens | 15,000 | $0.045 |
Gemini 3.5 Flash-Lite (audio input) | $0.3/1M audio tokens | 19,200 | $0.0058 |
Estimates use standard list prices and model-specific token rates. Google's Live guide uses about 25 audio tokens/sec; its general audio guide lists 32, while media-resolution docs also list 25. Gemini estimates cover new input only, excluding generated output and repeated Live session context. Check actual usage for billing. OpenAI's older transcription models are deprecated and retire Feb 26, 2027. Actual bills depend on your plan, region, and provider rounding.
Audio tokens FAQ
How many tokens is 1 minute of audio?
This calculator estimates 1,500 tokens per minute for Gemini 3.8 Live, using the Live guide's approximate 25 tokens per second. For audio-file input it retains the general audio guide's 32 tokens per second, or 1,920 per minute. Google's media-resolution docs also list 25, so use the selected endpoint's actual usage to reconcile costs. gpt-transcribe costs $0.0045 per audio minute.
Why is audio transcription billed per minute instead of tokens?
Transcription models like whisper-1 and gpt-4o-transcribe charge for audio duration because the work scales with recording length, not with how many words are spoken. Gemini's native audio models bill per audio token instead.
How much does it cost to transcribe 1 hour of audio?
The duration estimate is $0.27 with gpt-transcribe, $0.18 with gpt-4o-mini-transcribe, or $0.36 with whisper-1 and gpt-4o-transcribe. Gemini 3.5 Flash-Lite input alone is about $0.03 at the audio-file estimate used here; generated text costs extra. The older OpenAI models are deprecated and retire on February 26, 2027.
Does this calculator upload my audio?
No. If you pick a file, the browser only reads its duration metadata. The audio never leaves your device.
Which OpenAI transcription model should I use?
Use gpt-transcribe at $0.0045 per audio minute — it is OpenAI's current recommended model for recorded speech. whisper-1 and the gpt-4o transcription models (including the diarization variant) were deprecated on Aug 26, 2026 and will be removed from the API on Feb 26, 2027.
How many audio tokens per second does Gemini use?
Google's Live best-practices guide estimates about 25 tokens per second, while the general audio-understanding guide lists 32. The media-resolution guide also lists 25 for Gemini 3. This calculator distinguishes Live and audio-file estimates instead of promising one universal rate. Check your endpoint's usageMetadata for the actual billed count.