← All posts

How Much Does OpenAI's Whisper Actually Cost? Real Numbers from 3 Months

Subscription dictation apps justify their prices by making the underlying API cost feel like a black box. Here are three months of real OpenAI invoices from daily Mac dictation — so you can do the math yourself.


The Simple Answer

OpenAI charges $0.006 per minute of audio for Whisper transcription. That's the published rate, no tricks or tiers.

For most people who dictate regularly on a Mac, three months of usage costs between $6 and $20 total. That's not per month. That's the total for the quarter.

Subscription dictation apps typically charge $10–15 per month — $30–45 per quarter — for the same underlying transcription quality. The markup exists because they're running infrastructure, supporting teams, and building features. But if you want the transcription without the overhead, the API cost is what it actually costs.


Three Months of Real Numbers

Here are three consecutive months of OpenAI API invoices, using ParlaParla for daily Mac dictation:

Month 1 — Light use

  • Usage: ~15 min/day, 18 workdays
  • Total audio: ~270 minutes
  • OpenAI charge: $1.62
  • Notes: Lighter month, fewer long-form documents

Month 2 — Normal use

  • Usage: ~25 min/day, 21 workdays
  • Total audio: ~525 minutes
  • OpenAI charge: $3.15
  • Notes: Typical month with regular dictation workflow

Month 3 — Heavy use

  • Usage: ~40 min/day, 22 workdays
  • Total audio: ~880 minutes
  • AI polishing: enabled on most sessions
  • OpenAI charge: $5.28 (transcription) + $1.40 (GPT-4o mini for polishing) = $6.68
  • Notes: Heavy writing month, AI polishing used extensively

3-Month Total

  • Total OpenAI charges: $11.45
  • Equivalent subscription cost (Wispr Flow): $44.97
  • Savings: $33.52
  • Annualized savings at this usage: ~$115

What Affects Your Cost

There are three variables that determine your monthly OpenAI bill when using a BYOK dictation app:

1. How much you dictate

This is the biggest factor. The $0.006/minute rate is linear — every minute of audio costs the same. There's no volume discount, but there's also no minimum. If you only dictate 10 minutes on a given day, you pay $0.06.

Daily usage Monthly minutes Monthly cost Annual cost
5 min/day ~100 min $0.60 $7.20
15 min/day ~300 min $1.80 $21.60
30 min/day ~600 min $3.60 $43.20
60 min/day ~1,200 min $7.20 $86.40
120 min/day ~2,400 min $14.40 $172.80

For context: 120 minutes of dictation per day is roughly a full-time professional transcription workload. Most knowledge workers who use dictation occasionally are at 15–30 minutes per day.

2. Whether you use AI polishing

AI polishing — taking the raw Whisper transcription and running it through a language model to fix grammar, remove filler words, and improve flow — adds a separate cost.

ParlaParla uses OpenAI's GPT-4o mini model for polishing, which is priced at input/output token rates (not per-minute). For typical dictation sessions, the polishing cost adds roughly 10–30% on top of the transcription cost.

If your transcription for the month is $3.00 and you polish everything, expect to pay around $3.30–3.90 total. It's not significant at these usage levels.

If you don't need polishing — you prefer raw transcription and edit manually — you skip this cost entirely.

3. The language you're dictating in

The Whisper transcription rate is the same regardless of language: $0.006/minute. Whether you're dictating in English, Danish, German, or Spanish, the cost is identical. This is one area where the OpenAI API is simpler than subscription apps, which sometimes charge differently or restrict language support by plan tier.


How to Read Your OpenAI Invoice

OpenAI's billing dashboard breaks down costs by model. When using a BYOK dictation app, look for:

  • whisper-1 — this is the transcription model. Costs appear as "Audio" in the dashboard, priced per minute.
  • gpt-4o-mini (or similar) — if your app offers AI polishing, this appears separately as token usage.

The dashboard updates in near real-time, so you can check after a day of heavy dictation to see exactly what you spent. You can also set a hard usage limit in Settings → Limits — any amount you choose. Once you hit the limit, the API stops accepting requests until the next billing cycle. It's a simple guardrail if you're new to pay-per-use and want a ceiling.


When Does a Subscription Make Sense?

Let's be honest about the math here.

If you dictate 2+ hours per day, every workday, a subscription can start to look competitive. At 120 minutes/day for 22 workdays, your OpenAI bill would be around $14.40/month — roughly the price of a Wispr Flow subscription. At that level, a subscription isn't ripping you off.

For everyone who dictates less than 90 minutes per day — which is the vast majority of Mac users who use dictation at all — the BYOK model is significantly cheaper. Often by a factor of 3–5x.

There's also a second consideration: if you value a polished UI, automatic transcription history, and a team dedicated to building dictation-specific features on top of the core API, a subscription app may be worth the premium. You're paying for product development, not just API costs.

But if you want accurate transcription, no subscription, and direct control over your data — BYOK is the better choice.


How to Estimate Your Cost Before You Switch

You don't have to guess. Here's a simple way to estimate your monthly OpenAI cost before switching to a BYOK app:

  1. Track how long you spend dictating in a typical week. Even a rough estimate works — 10 minutes? 30? An hour?
  2. Multiply by 4.3 (average weeks in a month) to get monthly minutes.
  3. Multiply by $0.006 to get your monthly transcription cost.
  4. Add 20% if you plan to use AI polishing.

Example: 20 minutes/day × 5 days × 4.3 weeks = 430 minutes/month. 430 × $0.006 = $2.58/month. With polishing: ~$3.10/month.

Compare that to what you're currently paying for a subscription, and you have your answer.


The Bottom Line

Three months of daily dictation through the OpenAI Whisper API cost $11.45. The equivalent subscription would have cost $44.97. The transcription quality was identical — it's the same model either way.

The API cost of Whisper is low because transcription is a solved, commoditized problem. Subscription apps charge a premium for product development, support, and infrastructure. Whether that premium is worth it is a personal calculation. But you should make it with the actual numbers in front of you.

Now you have them.

Download ParlaParla

Use your own OpenAI key. Pay what you actually use.