The real cost of transcription services in 2026

The "price per minute" everyone compares is a misleading anchor. The true cost of transcription is dominated by what the rate is attached to: a per-seat SaaS subscription, a paid Workspace/Teams/Zoom license, human review labor, or your own engineering time. The headline rate is almost never the bill you actually pay.

Here is the short version. For the most common case (recurring meetings), a flat per-seat AI meeting tool is the best value, because per-minute API pricing looks cheap but ignores capture, recording, diarization, summary, storage, and your time. Human services only make sense for the small slice of work where 99%-plus legal-grade accuracy is non-negotiable. And "free" native captions in Meet, Teams, and Zoom usually carry a quiet license tax. Pick the bucket that matches how you transcribe, then compare inside that bucket.

The real question isn't "price per minute," it's "price per how you transcribe"

There is no single market rate for transcription. There are four different markets, each priced on a different unit, and comparing across them is the single biggest mistake people make.

1. Raw speech-to-text APIs, priced per minute or per hour of audio. You write the code, you store the output, you handle everything else.
2. Human transcription services, priced per audio minute, where a person (or a person plus AI draft) types and proofreads.
3. AI meeting tools, priced per seat per month, where minutes are bundled (and often unlimited) into a flat subscription.
4. Native platform captions, "free" inside Google Meet, Microsoft Teams, and Zoom, but gated behind paid licenses and manual activation.

A $0.006-per-minute API and a $20-per-seat subscription are not the same product priced differently. They are different products. Once you know your bucket, the comparison gets simple. Picking the wrong bucket is how people end up overpaying by 10x or, worse, building a homegrown pipeline that costs more in engineering time than any vendor would have charged.

What transcription actually costs in 2026 (verified rates by category)

Here are the real, current numbers, every one traced to a vendor pricing page as of June 2026.

$1.99/min

Rev human, 99%+ accuracy

$0.25/min

Rev AI, 95%+ accuracy

$0.006/min

OpenAI gpt-4o-transcribe

$0.15/hr

AssemblyAI Universal-2

300 min

Otter free monthly cap

15 mtgs

Scribbl free monthly cap

Human services. Rev charges $1.99 per audio minute for professional human transcription at 99%-plus accuracy with turnaround in 12 hours or less. GoTranscript advertises human work from roughly $1.02 to $1.58 per minute on its public cost estimator, with rush (6 to 12 hour) orders priced higher. That works out to roughly $60 to $120 for a one-hour recording, before verbatim or rush surcharges.

AI meeting tools (per seat). Otter.ai Pro is $16.99 per user per month, or $8.33 billed annually, with 1,200 transcription minutes per month and a 90-minute meeting cap. Otter Business is $30 per user per month, or $19.99 annually, with unlimited meetings and a 4-hour cap. Rev sells subscription tiers too: Essentials at $25.49 per seat per month billed yearly (5,000 AI minutes), Pro at $47.99 billed yearly (10,000 AI minutes). Scribbl is free for up to 10 meetings a month at unlimited length, and Pro is $20 per user per month, or $13 billed annually, with unlimited meetings.

Raw STT APIs (per minute or hour). AssemblyAI async speech-to-text is $0.15 per hour on Universal-2 and $0.21 per hour on Universal-3 Pro; streaming Universal-3 Pro starts at $0.45 per hour. OpenAI lists gpt-4o-transcribe at an estimated $0.006 per minute and gpt-4o-mini-transcribe at $0.003 per minute. New AssemblyAI users get $50 in free credits with no card. On paper these are the cheapest numbers in this whole article. They are also the most misleading, as the next section explains.

Native platform captions. Genuinely "free" only in the narrow sense covered below, and only Zoom's live captions clear that bar.

Human services

• Price: ~$1.02 to $1.99 per audio minute
• Accuracy: 99%+ (human proofed)
• Includes: a person typing and reviewing
• Best for: legal, medical, regulatory, published quotes
• Watch: rush and verbatim surcharges

AI meeting tools (per seat)

• Price: $0 to ~$30 per user per month
• Accuracy: ~90 to 95%
• Includes: capture, recording, diarization, summary, storage
• Best for: recurring meetings, most teams
• Watch: minute caps, video-retention expiry

Why per-minute API pricing is a trap for meeting transcription

Look at OpenAI's $0.006 per minute. A one-hour meeting costs about 36 cents in API spend. That sounds unbeatable until you remember what the API does not do.

The API turns audio into text. It does not capture the audio off your call. It does not record the meeting. It does not figure out who said what (that is diarization, often a separate charge or model). It does not summarize, pull action items, store the result anywhere searchable, or give you a UI to read it later. Every one of those is a line item you now own, either as more vendor spend or as engineering and operations time.

There is also a billing gotcha specific to streaming. AssemblyAI's real-time API bills for the time the WebSocket connection is open, not the audio you actually send. Idle time on an open connection counts. So a sloppy integration that holds connections open through silence quietly inflates a rate that looked like fractions of a cent.

APIs genuinely win in one scenario: a high-volume, programmatic pipeline where you are transcribing thousands of hours and already have the engineering to handle everything around the text. For a person or a team transcribing meetings, that math almost never works out. If you mostly need clean notes from calls, start with the best meeting transcription software rather than wiring up an API.

"Free" native transcription isn't free (Meet, Teams, Zoom)

This is the most common myth, and it costs people real money because they discover the catch only after they have built a workflow around it. Here is what "free" native transcription actually requires:

• Google Meet: native meeting transcription works only on Business Standard, Business Plus, Enterprise Standard, Enterprise Plus, and Education Plus editions. Someone in the meeting has to start it manually, and the transcript saves to the meeting organizer's Drive.
• Microsoft Teams: the live transcript lands in the organizer's OneDrive and is reachable from meeting chat and the Recap tab. Live translated transcription specifically requires Teams Premium.
• Zoom: auto-generated live captions have been free for all accounts since October 2021. But a saved transcript (not just on-screen captions) needs a paid plan plus cloud recording.

So Google Meet transcription is "free" only if your organization already pays for a Business Standard or higher Workspace edition, which is a per-seat license cost in its own right. Even then it is opt-in per meeting and owned by whoever organized the call, which is a problem if you want your own copy or you joined someone else's meeting. For a deeper walkthrough of the native path, see getting a Google Meet recording transcript.

The hidden costs here are admin setup, license seats, manual activation every single meeting, and organizer-only ownership. None of those show up on a "free" label, but all of them are real.

Human vs AI: when 8 to 10x the price is worth it (and when it's burning money)

The accuracy gap is real. Human transcription hits 99%-plus; good AI lands somewhere around 90 to 95%. The price gap is also real: roughly 8 to 10x. The question is whether your work actually needs that last few percent.

It does when the transcript is the deliverable and errors carry consequences: court and legal proceedings, medical records, regulatory filings, published verbatim quotes, and audio that is hostile to machines (heavy accents, many overlapping speakers, poor recording quality). In those cases the human premium is not a luxury, it is insurance.

It does not when the transcript is an internal artifact: meeting notes, action items, a searchable archive of your own calls. Here, 90 to 95% accuracy is plenty, and paying 8 to 10x for the difference is wasted spend. Nobody is going to subpoena your Tuesday standup.

The smart middle path for accuracy-critical work is hybrid: let AI produce a fast, cheap draft, then pay a human to clean up only the segments that matter. You get most of the speed savings and spend the human rate only where it earns its premium.

The cost drivers nobody quotes you up front

The headline number is the start of the bill, not the end. These are the multipliers that quietly inflate it:

• Turnaround. Standard pricing assumes you can wait. Rush orders (6 to 12 hours) cost more, sometimes a lot more.
• Verbatim vs clean read. Capturing every "um," false start, and stutter takes longer and costs more than a cleaned transcript.
• Audio quality and speaker count. Bad audio and many speakers slow human transcribers and lower AI accuracy, which can push you toward the more expensive option.
• Multilingual and translation. Non-English audio and translated output usually carry surcharges; Teams translated transcription, for example, needs Teams Premium.
• Caps and expiry on AI tools. Otter Basic caps at 300 minutes a month with a 30-minute-per-conversation limit. Rev Free is 45 AI minutes a month. And many tools expire your saved video or audio after a retention window, so "unlimited" sometimes means "unlimited until it disappears."

Always price the all-in cost for your actual usage pattern, not the per-unit rate on the homepage.

How to actually choose: a decision framework by use case

Two questions sort almost everyone into the right, cheapest option: how often do you transcribe, and do you need 99% legal-grade accuracy?

1. 1

   Need 99% legal-grade accuracy?

   If yes, use a human service (Rev, GoTranscript) and budget it as a per-deliverable line item. Stop here.

2. 2

   One-off or rare transcription?

   Use a free tier or pay per minute: Otter Basic, Zoom live captions, or a single-file human order. Don't buy a subscription.

3. 3

   Recurring meetings, just you?

   Use a flat per-seat AI meeting tool. A free tier (like Scribbl's 10 meetings a month) often means you pay $0.

4. 4

   A team of meeting-heavy users?

   Per-seat unlimited beats per-minute every time at volume. Compare per-seat AI tools, not API rates.

5. 5

   High-volume programmatic pipeline?

   Use a raw STT API, but price in capture, diarization, storage, and engineering before you commit.

The pattern: heavy meeting users blow past per-minute math fast, so flat unlimited per-seat wins. Light or one-off users are genuinely better off on a free tier or paying per minute. Accuracy-critical deliverables justify the human premium. And the API route only pays off at programmatic scale with engineering already in place.

If you want the cheapest-possible angle specifically, the companion piece on the best free transcription software breaks down the no-cost options by category. For more on what each platform's saved transcript involves, see Zoom meeting transcription.

Where Scribbl fits (and why a Google Meet user can pay $0)

For the most common case in this article, recurring Google Meet calls, the cheapest correct option is often a flat per-seat AI tool, and that is the case Scribbl is built for.

Scribbl records, transcribes, and summarizes your Google Meet calls from the browser with no bot joining the meeting. The free tier covers 10 meetings a month at unlimited length, so an individual recording their regular calls frequently pays nothing. Pro is a flat $13 to $20 per seat with unlimited minutes, which undercuts per-minute API math for anyone with more than a handful of meetings, and it does not require a paid Business Standard Workspace edition to get a usable Meet transcript the way native Meet transcription does.

For teams, the desktop app adds Zoom and Microsoft Teams support along with CRM, Slack, and Drive integrations, so you avoid paying a separate per-platform license tax on each tool. Compare the tiers on the pricing page.

To be clear about when Scribbl is not the answer: if you need a court-admissible verbatim transcript, use a human service and budget for it. If you are running a high-volume programmatic pipeline, use a raw STT API. For everyone whose real job is turning meetings into notes and action items, a flat per-seat tool is almost always the cheapest path that actually works.

Frequently asked questions

How much does it cost to transcribe one hour of audio?

It depends entirely on the bucket. A human service runs roughly $60 to $120 an hour (Rev at $1.99 per minute, GoTranscript from about $1.02 to $1.58). A raw STT API is cents (OpenAI gpt-4o-transcribe at $0.006 per minute is about 36 cents an hour; AssemblyAI from $0.15 an hour), but excludes everything around the text. An AI meeting tool bundles that hour into a flat per-seat subscription ($0 to about $30 a month) regardless of how many hours you run.

Is AI transcription accurate enough for business?

For internal use, yes. AI transcription typically hits 90 to 95% accuracy, which is fine for meeting notes, action items, and a searchable archive. It is not enough when the transcript itself is a legal, medical, or published deliverable, where 99%-plus human accuracy is worth the 8 to 10x premium. Match the accuracy to the stakes.

Is Google Meet, Teams, or Zoom transcription free?

Mostly no, despite the reputation. Google Meet native transcription requires a Business Standard or higher Workspace edition and a manual start, and the transcript belongs to the organizer. Teams stores the live transcript in the organizer's OneDrive, and translated transcription needs Teams Premium. Zoom live captions are genuinely free for all accounts, but a saved transcript needs a paid plan plus cloud recording.

Per-minute or per-seat: which is cheaper for me?

Estimate your monthly minutes. If you only transcribe occasionally, per-minute pricing or a free tier is cheaper. If you have recurring meetings, do the division: a $13-per-seat unlimited plan beats per-minute the moment you pass a few hours a month, and the gap widens fast for meeting-heavy users. Per-minute pricing punishes consistency.

What's the cheapest way to transcribe meetings in 2026?

For one person with recurring Google Meet calls, a free AI tool tier (such as Scribbl's 10 meetings a month) is genuinely $0 and includes recording and summaries. For occasional one-off files, a free SaaS tier or Zoom's free live captions costs nothing. The cheapest option is rarely the lowest headline rate; it is the one whose bundled cost matches how often you actually transcribe.