Transcription economics

Find the cheapest speech-to-text API for your audio volume

Turn per-minute, per-hour, and streaming rates into one accepted-audio-minute cost, then add the retries and fixed fees the headline rate leaves out.

One page · one buyer job

cheapest speech-to-text API

Checked Oct 21, 2026

Estimate the monthly cost of transcribing a known audio volume and compare published STT rates without mixing hourly, per-minute, and subscription meters.

Primary meter
Submitted audio minutes, not accepted minutes
Common hidden unit
Add-ons such as diarization, redaction, and keyterm prompting
Optimization
Batch pre-recorded work; reserve streaming for live audio
Quality gate
Same audio sample, language mix, and acceptance rubric

Speech-to-text cost calculator

Enter a rate from the exact provider, model, and processing mode you are evaluating. The estimate uses your inputs only and is not a provider quote.

Minutes of audio you need transcribed and accepted after review.
Convert an hourly rate by dividing by 60 before entering it here.
Accepted minutes divided by all submitted minutes in your own sample.
Minimum commitments, storage, diarization add-ons, or another known fixed amount.
Reset example

Planning result

Review the example inputs, replace the rate with the exact current provider rate, and calculate a workload. APIDir does not send these values to a provider.

Token, video, audio, image, and 3D API workloads passing through normalized cost meters
AI concept diagramConcept diagram: each workload keeps its native billing unit before APIDir calculates a comparable planning total.

Start from the audio you must accept

Transcription quotes are usually expressed per minute or per hour of submitted audio, while a product needs minutes it can actually use. The gap is retries: noisy recordings, overlapping speakers, unsupported languages, and failed requests all consume the same meter. This calculator divides accepted minutes by your own acceptance rate before applying the published rate, so the estimate reflects the audio you must buy rather than the audio you hope to keep.

Use a representative sample across short and long recordings, clean and noisy rooms, and every language in the product. Measure word error rate and, more importantly, whether the transcript passes the downstream task: search indexing, summarization, compliance review, or captions. A cheaper model that fails that task is not cheaper per accepted minute.

Published rates normalized to one minute

The table converts each first-party rate to USD per audio minute using arithmetic only. It does not assume the models are equally accurate, and it does not treat a promotional streaming rate as a permanent price.

These are the published rates on 2026-10-21. Providers change prices, model names, and promotional terms; recheck the linked source before committing volume.

Provider and modelPublished ratePer audio minuteNotes
OpenAI gpt-4o-transcribe$0.006 / minute$0.006Listed transcription model
OpenAI Whisper$0.006 / minute$0.006Listed as Whisper
OpenAI gpt-4o-mini-transcribe$0.003 / minute$0.003Lower published rate
Deepgram Nova-3 pre-recorded, monolingual$0.0043 / minute$0.0043Pay-as-you-go; $200 free credit advertised
Deepgram Nova-3 pre-recorded, multilingual$0.0052 / minute$0.0052Pay-as-you-go
AssemblyAI Universal-3.5 Pro$0.21 / hour$0.0035Pre-recorded; $50 free credits advertised
AssemblyAI Universal-2$0.15 / hour$0.0025Pre-recorded

What the per-minute meter leaves out

A base transcription rate rarely includes every feature a workflow needs. Deepgram publishes separate per-minute add-on rates for redaction, keyterm prompting, entity detection, and streaming speaker diarization, while pre-recorded diarization is listed as included. AssemblyAI and OpenAI price additional products separately. Add those meters to the fixed-cost field or model them as a second calculation.

Streaming is a different service objective from pre-recorded transcription. Live audio is billed by connection time at rates that can exceed the pre-recorded rate, and the connection may stay open while nobody speaks. Do not use a streaming price to estimate a batch archive job, and do not use a pre-recorded rate to budget a live captions feature.

  • Count submitted minutes, including failed and retried requests.
  • Model add-on meters separately from the base transcription rate.
  • Keep pre-recorded and streaming workloads in separate estimates.
  • Recheck free-credit terms: they reduce the first invoice, not the unit rate.
  • Verify language coverage and speaker diarization on your own recordings.

Common comparison mistakes

  • Dividing an hourly rate by the wrong number or comparing it directly with a per-minute rate.
  • Budgeting accepted minutes while the provider bills submitted minutes.
  • Treating a promotional streaming rate as the permanent pre-recorded price.
  • Ignoring diarization, redaction, keyterm, and entity add-ons.
  • Assuming a free credit makes the ongoing unit cost lower.

Questions developers ask

Which speech-to-text API is cheapest?

On published base rates alone, AssemblyAI Universal-2 at $0.15 per hour is $0.0025 per minute, below Deepgram Nova-3 pre-recorded at $0.0043 and OpenAI at $0.003 to $0.006. The cheapest viable route is the one that passes your language, accuracy, and feature requirements at the lowest cost per accepted minute.

How do I convert an hourly transcription rate?

Divide the hourly rate by 60. For example, $0.21 per hour is $0.0035 per minute and $0.15 per hour is $0.0025 per minute. Enter the per-minute value in the calculator.

Should I use a free credit in the estimate?

Treat free credits as a first-invoice reduction, not a permanent discount. The calculator uses the ongoing published rate so the estimate remains valid after the credit is consumed.

Why does the calculator ask for an acceptance rate?

Providers bill submitted audio, but a product needs usable transcripts. The acceptance rate turns accepted minutes into the billable minutes you must purchase, including retries.

First-party boundaries

Official sources to recheck

These links define meters and product boundaries. APIDir does not freeze their current rates inside the planning calculator.

How APIDir handles evidence →
  • OpenAI API pricingOfficial transcription rates for Whisper, gpt-4o-transcribe, and gpt-4o-mini-transcribe.Checked Oct 21, 2026
  • Deepgram pricingOfficial pre-recorded, streaming, add-on, and free-credit terms.Checked Oct 21, 2026
  • AssemblyAI pricingOfficial pre-recorded and streaming rates plus free-credit terms.Checked Oct 21, 2026

APIDir Delta · controlled rollout

Join the source-linked pricing update waitlist.

Confirm your address to join the waitlist. Editorial delivery starts only after sender, suppression, and evidence checks are active.