Prices verified May 14, 2026 · Page updated October 4, 2026

How Much Does Transcription Cost in 2026? A Verified Price Reference

By VexaScribe Editorial · Published May 14, 2026 · Updated October 4, 2026 · Competitor prices verified against their public pricing pages May 14, 2026

Transcription costs range from $0 (self-hosted Whisper, in licence terms) to $1.99 per minute (human transcription via Rev). AI APIs cost $0.15–$1.44 per audio hour: AssemblyAI is the cheapest at $0.0025/min ($0.15/hr), OpenAI's Whisper API $0.36/hr, Deepgram Nova-3 $0.46/hr, AWS Transcribe Tier 1 $1.44/hr. Consumer AI apps charge $2–$25/month, which works out to $0.20–$1.60 per audio hour depending on how much of your tier you use: VexaScribe runs from $2/mo Starter (200 min, $0.60/hr effective) to $20/mo Studio (6,000 min, $0.20/hr effective), Otter Pro is $8.33/mo annual for 1,200 minutes, Descript Hobbyist $16/mo for 10 hours. Human transcription costs $90–$119 per audio hour — Rev starts at $1.99/min, Scribie at $1.50/min. That makes Rev's human rate 796× AssemblyAI's, because a person needs roughly four to six hours to transcribe one hour of audio and the price has to cover their working hours rather than your audio hours. Below: a calculator for your own volume, cost by method, cost by volume from 5 to 100 hours a month, the costs that sit outside the sticker price, and what each method genuinely suits.

Key takeaways

Cheapest-in-category claims below refer to the 14 options priced on this page, at their public rates on May 14, 2026. They are not a survey of every service in the market.

  • →Cheapest API: AssemblyAI at $0.15/hour ($0.0025/min PAYG, no subscription).
  • →Cheapest consumer app: VexaScribe — $2/mo Starter (200 min) scaling to $20/mo Studio (6,000 min). Effective per-hour drops from $0.60 at Starter to $0.20 at Studio, which lands in the same range as mid-tier APIs without the integration work.
  • →No licence cost: Self-hosted Whisper. The model is free; the GPU, setup and maintenance are not — see the fully-loaded figure below before assuming $0.
  • →Free and genuinely useful: YouTube auto-captions, if you are uploading to YouTube anyway. English-primary, and you cannot take the captions anywhere else easily.
  • →Human transcription: $1.99/min via Rev, $1.50/min via Scribie — $90–$119 per audio hour, or 796× the cheapest API rate.
  • →Subscription sweet spot: $5–$10/mo tier covers 1,000–2,500 minutes/mo across most consumer apps.
  • →Hidden cost #1: Per-seat pricing — Otter Business at $19.99/seat × 5 people = $99.95/mo, not $19.99.
  • →2026 market shift: Rev moved from per-minute PAYG to subscription-only AI ($25.49–$47.99/seat/mo). Human transcription still PAYG at $1.99/min.

Cost calculator: your monthly transcription bill

Enter how many minutes of audio you transcribe each month. The calculator ranks 13 options from cheapest to most expensive at your volume. Competitor prices verified May 14, 2026; VexaScribe tiers current as of October 4, 2026.

minutes (10.0 hours/month)

Cheapest 3 options at 600 min/mo

#1 · Self-hosted WhisperSelf-hosted
Free forever; requires GPU + Python setup
$0.00
/month
#2 · AssemblyAI Universal-2API
PAYG at $0.0025/min; engineering time required
$1.50
/month
#3 · OpenAI whisper-1 APIAPI
PAYG at $0.006/min; engineering time required
$3.60
/month
See all 12 options ranked
RankOptionCategoryMonthlyPer minute
1Self-hosted WhisperSelf-hosted$0.00$0.0000
2AssemblyAI Universal-2API$1.50$0.0025
3OpenAI whisper-1 APIAPI$3.60$0.0060
4Deepgram Nova-3API$4.62$0.0077
5VexaScribe BasicConsumer app$5.00$0.0083
6Otter Pro (annual)Consumer app$8.33$0.0139
7VexaScribe ProConsumer app$10.00$0.0167
8AWS Transcribe (Tier 1)API$14.40$0.0240
9Descript HobbyistConsumer app$16.00$0.0267
10VexaScribe StudioConsumer app$20.00$0.0333
11Rev AI subscription (Essentials)Consumer app$25.49$0.0425
12Rev Human transcriptionHuman$1194.00$1.9900

Competitor prices verified May 14, 2026 against each vendor's public pricing page; check the vendor before committing. Deliberately excluded, because they depend on your situation rather than your volume: overage rates, per-seat team pricing, annual-versus-monthly differences, and the self-hosting costs that sit outside the $0 licence — all covered in the section on costs that aren't on the pricing page.

Quick answer: cost per minute & cost per hour

Direct prices across every transcription method, sorted cheapest to most expensive. All prices verified against vendor pricing pages on May 14, 2026.

MethodCost per minuteCost per hour
Self-hosted Whisper$0$0
AssemblyAI Universal-2 (cheapest API)$0.0025$0.15
OpenAI whisper-1 API$0.006$0.36
Deepgram Nova-3$0.0077$0.46
VexaScribe Studio (cheapest consumer app at scale)$0.0033$0.20
Most consumer apps (effective)$0.01–$0.03$0.60–$1.80
AWS Transcribe (Tier 1)$0.024$1.44
Rev AI subscription (per minute amortized)~$0.005~$0.31
Human transcription (Rev)$1.99$119

How to read this table: "Cost per minute" and "cost per hour" are mathematically the same number (×60). Most tools quote one or the other on their pricing page — we've normalized both so you can compare directly. Consumer app rates are "effective" — your actual cost depends on your subscription tier and usage, not a true per-minute rate.

4 Transcription Methods Compared

Every transcription option in 2026 fits one of four methods. Understanding the methods explains the 800× price difference between the cheapest and most expensive options.

Method 1: Self-hosted AI (Whisper, pyannote)

$0 licence · Unlimited use · Your hardware, your maintenance

Run the open-source Whisper Large-v3 model on your own hardware. The model weights cost nothing; the compute does — a single A10 or consumer RTX card is enough for personal volumes.

On accuracy, the honest version: several consumer tools and OpenAI's own whisper-1 endpoint run the same Whisper weights you would be running, so output there is broadly comparable. AssemblyAI and Deepgram run their own models, and we have not measured any of them against each other — so treat self-hosting as a cost decision, not an accuracy upgrade.

Best for: developers comfortable with Python, on-premises or air-gapped requirements, or processing thousands of hours a month where per-minute commercial pricing stops making sense. Before assuming $0, read the fully-loaded figure in the hidden-costs section below — the licence is free, the total is not.

Method 2: AI APIs (AssemblyAI, Deepgram, OpenAI, AWS, Google)

$0.15–$1.44 per hour · PAYG billing · Engineering integration required

Commercial APIs that wrap Whisper or their own in-house models with managed infrastructure, SDKs, and SLAs. Cheapest here is AssemblyAI at $0.0025/min. The enterprise clouds (AWS, Google) charge roughly ten times that at Tier 1, and what you get for the difference is compliance posture — HIPAA, SOC 2 — rather than better transcripts. See our 12-API comparison for full vendor details.

Best for: engineering teams building transcription into their own product. The crossover against a consumer app is about engineering time rather than a fixed volume — at low volume, the hours spent building and maintaining the integration cost more than the per-hour saving returns.

Method 3: Consumer AI apps (VexaScribe, Otter, Descript, TurboScribe, Rev AI)

$2–$25/mo subscription · Web UI · No code

UI-first tools with web dashboards, file upload, transcript editing, and downstream features (summaries, subtitles, integrations). Effective per-hour rates run higher than raw API rates — $0.20 to $1.60 across the tiers priced here, against $0.15 to $1.44 for the APIs — and what the difference buys is that you do not have to build anything. 2026 market shift: Rev moved from per-minute PAYG to a subscription-only AI tier ($25.49–$47.99/seat/mo).

Best for: end-users with regular transcription needs who want a UI and no code. Sub-tier choice depends on volume: $2/mo for occasional use, $5–$10/mo for regular use, $20+/mo for high volume.

Method 4: Human transcription (Rev, Scribie, Verbit)

$1.50–$1.99/min standard · $3–$5/min legal-grade · 12–24 hour turnaround

Trained human transcriptionists handle overlapping speech, accents, technical jargon, and noise conditions that defeat every AI tool. Court-admissible. NDA-protected. The premium is arithmetic: $1.99/min × 60 = $119 per audio hour, against $0.15–$1.44 for the APIs — and it is paying for four to six working hours per audio hour, not an inflated margin.

Best for: legal depositions, peer-reviewed academic research, broadcast deliverables under regulatory scrutiny, and audio with heavy crosstalk where AI fails. For everyday content, the hybrid approach — AI draft, human correction on the passages that matter — costs a fraction of full human transcription, scaling with how much you have reviewed.

Cost by method: full pricing table

14 tools across all four methods with entry-tier pricing, best-value-tier pricing, and effective per-hour cost. Each row was verified against the vendor's pricing page on May 14, 2026.

ToolMethodEntry tierBest-value tierPer hour
Self-hosted WhisperOpen-source$0$0$0
AssemblyAI Universal-2APIPAYG $0.0025/minPAYG$0.15
OpenAI whisper-1APIPAYG $0.006/minPAYG$0.36
OpenAI gpt-4o-transcribeAPIPAYG $0.006/minPAYG$0.36
Deepgram Nova-3APIPAYG $0.0077/minVolume discount$0.46
AWS TranscribeAPI$0.024/min Tier 1Volume tier (>250K min)$1.44 (Tier 1)
Google Cloud Speech-to-TextAPI$0.024/minVolume tier$1.44
VexaScribeConsumer app$2/mo Starter (200 min)$20/mo Studio (6,000 min)$0.20–$0.60
Otter ProConsumer app$8.33/mo annual (1,200 min)$8.33/mo annual$0.42
Descript HobbyistConsumer app$16/mo (10 hrs)$24/mo Creator (30 hrs)$0.80
TurboScribe UnlimitedConsumer app$10/mo unlimited$10/mo unlimitedVariable
Rev AI (subscription)AI tier$25.49/seat/mo annual (5,000 min)Essentials annual$0.31
Rev HumanHuman$1.99/min startingPAYG$119
Scribie HumanHuman$1.50/min standardStandard 5-day$90

How to read "per hour": For PAYG (APIs), this is the literal rate × 60. For subscriptions, it's computed as monthly tier ÷ included minutes, then ×60. "Variable" on TurboScribe Unlimited means the per-hour cost drops with usage (at 50 hours/month, $10 ÷ 50 = $0.20/hour effective).

Cost by volume: 5 / 20 / 50 / 100 hours per month

The single most useful table for choosing a tool. Same options across four volume levels — pick your monthly usage and read across. Prices include subscription floor where applicable.

Tool5 hrs/mo20 hrs/mo50 hrs/mo100 hrs/mo
Self-hosted Whisper$0$0$0$0
AssemblyAI PAYG$0.75$3.00$7.50$15.00
OpenAI whisper-1 API$1.80$7.20$18.00$36.00
Deepgram Nova-3$2.31$9.24$23.10$46.20
AWS Transcribe Tier 1$7.20$28.80$72.00$144.00
VexaScribe Starter/Basic/Pro/Studio$2 Starter$5 Basic$10 Pro$20 Studio
Otter Pro (annual)$8.33$8.33 (cap)Need Business $19.99Need Business $19.99
Descript$16 Hobbyist$24 Creator$50 Business + overageCustom Enterprise
TurboScribe Unlimited$10$10$10$10
Rev AI Essentials$25.49$25.49 (~83 hr cap)$25.49 (cap exceeded)Need Pro $47.99
Rev Human ($1.99/min)$597$2,388$5,970$11,940

Reading guide: At 5 hours/mo, almost everything is cheap — pick on UX. At 20 hours/mo, AI APIs and entry-tier subscriptions tie. At 50 hours/mo, flat-rate plans (VexaScribe Pro $10, TurboScribe $10) crush per-hour pricing. At 100 hours/mo, VexaScribe Studio ($20) or self-hosted Whisper are the only practical AI choices. Rev Human at any volume becomes prohibitive above 1 hour/month.

AI vs Human: when is each worth it?

Rev's human tier costs 796× the cheapest API rate. That gap is the single biggest number on this page, so it is worth knowing where it comes from before deciding which side of it you belong on. This section covers the cost case; for the quality argument in depth, see AI vs human transcription.

Two different hours, and the confusion they cause

Services quote per audio hour — one hour of recording, whatever it takes to produce. People are paid per working hour — an hour of someone's time. These are not the same unit, and transcription is one of the few tasks where they diverge sharply.

Transcribing speech takes a competent person roughly four to six times the length of the audio: stopping, rewinding, resolving crosstalk, checking spellings. So one audio hour is most of a working day's usable output. At $119 per audio hour, Rev is not paying anyone $119 an hour — it is covering four to six working hours, plus review and overheads, out of that figure.

This matters in two directions. If you are buying, a quote per audio hour is the only one you can compare across vendors, and a quote per working hour will almost always look cheaper than it is. If you are a transcriptionist pricing your own work, the realtime multiple is the whole negotiation — an audio-hour rate that ignores it prices your day at a fraction of your hourly rate.

Use AI when:

  • ✓A transcript you will read and edit is acceptable — podcasts, interviews, internal meetings
  • ✓You have time to proofread proper nouns (names, brands, technical terms)
  • ✓Volume exceeds 1 hour/month (human cost compounds fast)
  • ✓You can edit the transcript yourself (or already plan to)
  • ✓Audio is clean — single speaker or 2–4 cooperative speakers

Use Human when:

  • →Legal depositions (court-admissible record required)
  • →Peer-reviewed academic research (publication-grade accuracy)
  • →Broadcast deliverables (FCC/regulatory standards)
  • →Heavy overlap, crosstalk, or debate audio (AI fails here)
  • →Specialized vocabulary AI can't learn (rare medical, legal, technical terms)
  • →Final-draft content where editing time costs more than the human transcription

The hybrid approach

Transcribe with AI first, then pay a person to correct only the passages that matter — usually the quotes you intend to publish, not the whole recording. You keep the AI per-audio-hour rate across the full length and buy human attention by the working hour for a fraction of it.

What makes this work is the unit switch from the section above: correcting a draft is a working-hour task measured in minutes per quote, while transcribing from scratch is a working-hour task measured in multiples of the audio length. Your saving depends on how much of the transcript genuinely needs a human, so the honest answer is that it scales with your review scope — a few quotes from a long interview is a large saving, a full verbatim check is close to none.

The costs that aren't on the pricing page

Seven categories that change the real bill. The first six apply to paid services; the seventh applies if you were planning to self-host because the licence is free.

1. Per-seat pricing

A 5-person team on Otter Business at $19.99/seat = $99.95/mo total, not $19.99. Per-seat pricing scales linearly — for teams, factor headcount × seat price. Flat-rate alternatives (TurboScribe $10/mo unlimited, VexaScribe $35/mo team plan) are 3-10× cheaper at 5+ seats.

2. Overage rates

Going 100 minutes over your monthly plan often costs $0.10–$0.50/min — sometimes more than upgrading to the next tier would have. Always set usage alerts at 80% and 95% of your plan limit. Some tools (Rev AI Pro) don't allow overage and simply stop processing until next month.

3. Feature gating

Speaker diarization, AI summaries, custom vocabulary, and API access are often locked to higher tiers. A $2/mo plan that excludes diarization isn't useful for multi-speaker audio. Always confirm the features you need are on your tier before committing.

4. Annual vs monthly billing lock-in

Annual billing is cheaper per month everywhere it is offered, and it commits you. Otter Pro is the clearest worked example at the prices verified here: $16.99 monthly, or $8.33/mo on the annual plan. Run it 12 months and annual costs $99.96 against $203.88 — a saving of just over half. Cancel after four months and the annual plan has still charged you for the year, while four monthly payments come to $67.96 and stop. The question is not which rate is lower, it is whether you will still be using this tool in a year.

5. Format export fees

A few tools charge extra for SRT/VTT export above their base tier. Most modern tools include all formats free — but verify before subscribing. Sonix bills SRT/VTT export as part of subscription but PAYG users pay per export.

6. API rate limits and retry costs

Streaming APIs that disconnect (WebSocket timeouts, mobile network changes) often require you to retry, and each retry consumes more minutes. Implement exponential backoff and cap retry attempts, because a retry loop bills every attempt.

7. What self-hosting costs that isn't the licence

Whisper's weights are free. The rest is not: GPU time (rented by the hour, or bought once and then idle between jobs), the engineer-hours to build upload handling, queueing, retries and output formatting, the monitoring to notice when it breaks, storage and bandwidth for the media, and the re-work when a new model version ships. None of these appear in a $0 figure, and at low volume they dominate it — a few hours of engineering time exceeds a year of any subscription on this page. Self-hosting wins on cost at sustained high volume, where the per-minute rates compound and the fixed costs amortise. Price the whole list against a managed service before committing, not just the licence.

What this list looks like for VexaScribe

Since we publish this page, here is the same audit applied to us. Minutes are the only thing billed: summaries, chat and translation run on an existing transcript and consume no additional minutes. One audio minute is one transcription minute. All four tiers include speaker labels and every export format, so category 3 above — feature gating — does not apply. There is no overage charge; a plan stops at its cap and waits for the next cycle or an upgrade.

The genuine limitation is that cap. Studio tops out at 6,000 minutes a month, so above roughly 100 audio hours we are not the cheapest answer on this page and the calculator above will say so — at that volume, a PAYG API or self-hosting wins. Premium-quality transcription is also opt-in per upload and consumes two minutes per audio minute instead of one, which the per-hour figures in our rows assume you are not using.

Cost by use case

Different use cases land on different tools. Concrete monthly cost estimates for the five most common scenarios.

Podcast transcription cost

For a weekly 30-minute podcast: 4 episodes × 30 min = 120 minutes/month (2 hours).

  • →AI per-episode cost: $0.075 (AssemblyAI), $0.18 (OpenAI), $0.23 (Deepgram), ~$0.10 (VexaScribe amortized)
  • →Human per-episode cost: $60 (Rev Human at $1.99/min × 30 min)
  • →Annual cost (weekly 30-min podcast): $24/year on VexaScribe Starter, $20/year on TurboScribe Unlimited, or $36/year on OpenAI API
  • →Anchor recommendation: VexaScribe Starter ($2/mo for 200 min) covers a weekly 30-min podcast for $24/year
See full podcast tool comparison →

Interview transcription cost

For thesis-scale research: 20 interviews × 60 min = 1,200 minutes (20 hours).

  • →AI total cost: $3.00 (AssemblyAI), $7.20 (OpenAI), $9.24 (Deepgram), $5/month on VexaScribe Basic
  • →Human total cost: $2,388 (Rev Human at $1.99/min × 1,200 min)
  • →Hybrid route: AI all 20 interviews for ~$5–$10 total, then pay a reviewer by the hour to clean up only the 3-5 you will quote directly — you buy working hours for a handful of passages instead of for 20 audio hours
  • →For journalists doing 2 interviews/week (~50 hours/year): VexaScribe Basic at $5/month = $60/year covers the entire workload
See speaker diarization tools comparison →

Video subtitles cost

For YouTube creators: 4 videos × 15 min = 60 minutes/month (1 hour).

  • →Free option: YouTube auto-captions ($0, English-primary, and the captions stay inside YouTube)
  • →AI dedicated SRT generation: $0.15/month (AssemblyAI), $0.36 (OpenAI), or $2/month on VexaScribe Starter, which adds bulk upload and an editor
  • →For high-volume creators (50+ videos/mo): VexaScribe Studio at $20/mo, or self-hosted Whisper if you have the GPU
  • →Why bother at all: social feeds autoplay muted, so uncaptioned video is silent video. Any of the paid options above costs less per month than one freelance caption job
Try VexaScribe's SRT generator free →

Meeting transcription cost

For a 5-person team with 1 hour of meetings per person per workday: 5 × 1 × 20 = 100 hours/month total.

  • →Flat-rate team plans win at this scale: VexaScribe Studio $20/mo flat, TurboScribe Unlimited $10/mo flat
  • →Per-seat pricing scales painfully: Otter Business $19.99/seat × 5 = $99.95/mo, Fireflies Pro $19/seat × 5 = $95/mo
  • →AI API direct: AssemblyAI PAYG at 100 hours = $15.00/month (cheapest), but requires engineering integration with your meeting recording workflow
  • →Hidden cost: meeting transcription requires bot capability OR upload workflow — pure API doesn't capture live meetings without recording infrastructure
See speaker diarization tools comparison →

Free options that actually work

"Free" usually means time-limited, file-limited, or platform-coupled. Five options that genuinely deliver — honest about each catch.

ToolLimitEngineCatchVerdict
Self-hosted WhisperUnlimited, no licence costWhisper Large-v3, run by youRequires a GPU and Python setupLowest cost at high volume, if you have the skills
YouTube auto-captionsUnlimited if the video is on YouTubeGoogle's own model, English-primaryCaptions stay inside YouTubeFree and adequate for YouTube-only workflows
VexaScribe 30-min trial30 minutes, one-time, no cardWhisper Large-v3 on the Standard tierOne-time trial, not a recurring free tierEnough to test a real file end to end
TurboScribe free3 files/day, 30-min limit eachWhisper-basedDaily file count, not minutesWorks for occasional single files
Otter free300 min/mo, 30-min recording capOtter's own model, English-primaryPer-recording cap bites on long meetingsThe most generous recurring free tier here

Truly unlimited free use requires self-hosting Whisper or hosting on YouTube. For a one-off standalone SRT file from any audio or video, no account needed for the first 30 minutes — try VexaScribe's SRT generator.

Frequently Asked Questions

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

One hour of audio costs $0 to $119, depending on method. Self-hosted Whisper: $0 in licence terms, though not in compute or setup time. AssemblyAI API: $0.15/hour. OpenAI Whisper API: $0.36/hour. Deepgram Nova-3: $0.46/hour. AWS Transcribe Tier 1: $1.44/hour. Consumer apps: $0.20–$1.60/hour effective, depending on how much of your subscription tier you actually use. Rev Human transcription: $119/hour, starting at $1.99/min. For podcasts, interviews and internal meetings, AI is the normal choice; the human premium is justified for legal depositions, peer-reviewed research, or broadcast deliverables. Prices verified May 14, 2026 against each vendor's public pricing page.

What is the cheapest transcription service in 2026?

Of the 14 options priced on this page, at their public rates on May 14, 2026: lowest cost with no card is self-hosted Whisper, free in licence terms but requiring a GPU and Python. Cheapest commercial API is AssemblyAI at $0.0025/min ($0.15/hour) pay-as-you-go. Lowest consumer-app entry tier is VexaScribe at $2/month for 200 minutes ($0.60/hr effective), falling to $0.20/hr effective at $20/month for 6,000 minutes. Best free tier for ongoing use is YouTube auto-captions if you are uploading to YouTube anyway, English-primary, with the captions staying inside YouTube. TurboScribe's free tier allows 3 files per day at 30 minutes each.

Why is human transcription so much more expensive than AI?

Rev's human rate of $1.99/min is 796× AssemblyAI's $0.0025/min — the gap is arithmetic, not marketing. Three things drive it. (1) Transcription takes a person roughly 4 to 6 times the audio length, so one audio hour is most of a working day's output, and the price has to cover working hours rather than audio hours. (2) Human transcriptionists handle overlapping speech, crosstalk and strong accents that remain genuinely hard for every AI tool. (3) A human transcript can be certified for court; an AI transcript cannot. For podcasts, interviews and internal meetings, AI output is normally the sensible starting point. The human premium is justified for legal depositions, peer-reviewed research, or broadcast deliverables under regulatory review.

Are there any truly free transcription services?

Five genuinely free options work for specific use cases. (1) Self-hosted Whisper — no licence cost, unlimited, requires Python and a GPU. (2) YouTube auto-captions — free and English-primary, useful if you are uploading to YouTube anyway, though the captions stay inside YouTube. (3) VexaScribe free 30-minute trial — one-time, no card required. (4) TurboScribe free tier — 3 files/day at 30 min each. (5) Otter free tier — 300 min/month, English-primary, 30-min cap per recording. Truly unlimited free requires self-hosting; everything else has time, file, or quantity limits.

How much does it cost to transcribe a 30-minute podcast?

A 30-minute podcast episode costs $0–$60 depending on method. Free: YouTube auto-captions (if uploading), self-hosted Whisper. AI APIs: $0.075 (AssemblyAI), $0.18 (OpenAI), $0.23 (Deepgram). Consumer apps amortized over a $2/mo subscription: roughly $0.50–$2 per episode. Rev Human: $60 ($1.99/min × 30 min). For weekly podcasters publishing 4 episodes/month (~2 hours total), VexaScribe Starter at $2/month covers the year for $24 total — cheaper than a single hour of human transcription.

Is paying $1.99/min for Rev Human worth it?

Worth it for: legal depositions (court-admissible record), peer-reviewed academic research, broadcast deliverables under regulatory standards, or audio with heavy overlap and crosstalk that defeats AI. Not worth it for: podcasts, interview drafts, internal meetings, training videos, or anything you were going to edit anyway. The middle path is to transcribe with AI first, then have a person correct only the passages that matter — the quotes you will actually publish. That keeps the per-audio-hour AI cost and buys human attention by the working hour instead of across the whole recording, which is where the 796× gap comes from.

What's the difference between AI transcription tools and AI APIs?

Some run the same underlying models — OpenAI's whisper-1 endpoint and several consumer tools both run Whisper — while AssemblyAI and Deepgram run their own. The real difference is cost and integration. AI APIs (AssemblyAI $0.15/hr, OpenAI $0.36/hr, Deepgram $0.46/hr) require engineering: you write code to upload files and process responses. Consumer apps (VexaScribe $0.20–$0.60/hr, Otter $0.42/hr, Descript $0.80–$1.60/hr) cost more per hour but include a UI, file management and a dashboard, with no code. The crossover is about engineering time rather than a fixed volume: below a few hundred hours a month, the hours spent building against an API typically cost more than the per-hour saving returns.

Do transcription services charge per minute or per word?

Per minute, almost universally. Both AI APIs and human services bill by audio duration, not transcript length. Per-second billing (AWS Transcribe) and per-word billing are both rare. The standard unit is the audio minute — silence and pauses count. This matters because a 30-minute interview with 5 minutes of dead air still costs the same as 30 minutes of dense speech. If your recordings carry long silences, as lecture captures and paused interviews often do, trimming them before upload reduces the billed duration by exactly the length you remove.

What's the cheapest way to transcribe video content?

Cheapest path for video: extract the audio track (use ffmpeg or any video editor) and submit it to an AI tool. For YouTube videos: use YouTube's free auto-captions if you're uploading anyway. For dedicated SRT subtitle files: VexaScribe at $0.01/min (cheapest consumer app with SRT export), or self-hosted Whisper for unlimited free use. Most consumer apps and APIs accept video files directly (MP4, MOV, MKV) — they extract audio internally. You don't need to convert to MP3 first unless your upload bandwidth is slow.

How can I estimate my monthly transcription cost?

Use the calculator above on this page. Or do the math: take your monthly audio hours × the per-hour cost of your chosen tool. Example: a podcaster recording 4 hours/month picks VexaScribe Basic ($5/mo for 1,000 min) — total cost $5/month. A journalist conducting 20 interviews of 30 min each (10 hours total) on AssemblyAI PAYG = 10 × $0.15 = $1.50. A 5-person team using Otter Business at $19.99/seat = $99.95/month flat. For high-volume use (50+ hours/month), VexaScribe Studio ($20/mo flat) or self-hosted Whisper become the cheapest options.

Methodology & disclosure

Verification window. Competitor prices, free-tier limits and plan structures on this page were checked against each vendor's public pricing page between May 8 and May 14, 2026. That is the date in the badge at the top, and it is the date that matters for every figure in the tables below the calculator. Transcription pricing moves — by the time you read this, individual rates may have changed, and the vendor's own page is always the authority. Our own four plan prices are current as of October 4, 2026.

What the October 2026 update did and did not do. This revision corrected the page's structure and removed claims it could not support: a headline API range of "$0.15–$2.04/hour" when no figure above $1.44 appears in any table, five accuracy percentages with no source behind them, a viewer-retention statistic, and a set of freelance-rate and savings figures that were never measured. It also added the self-hosting cost section and the audio-hour versus working-hour explanation. It did not re-check a single competitor price. Those still carry their May date, honestly labelled, because re-verifying roughly thirty figures means opening thirty vendor pricing pages — and claiming a verification that did not happen is the failure this update existed to fix.

Methodology. We use vendor list pricing only — no negotiated enterprise discounts, no beta tier pricing, no insider rates. Per-minute and per-hour rates are computed from monthly tier ÷ included minutes, then ×60. PAYG rates are taken directly from vendor pricing pages. For Rev specifically, we document the 2026 shift from per-minute PAYG to subscription-only AI tier.

What we ignored. Marketing claims of "industry leading" without published benchmarks, accuracy claims without dataset disclosure, vendor-paid analyst reports, "starting from" prices that conceal real entry costs.

Conflict of interest. This page is published by VexaScribe. VexaScribe appears in the tables where its published prices place it — the lowest entry tier among the consumer apps priced here, at $2/mo, and not the cheapest option overall, which is self-hosting followed by the PAYG APIs. We make no accuracy or feature claim for VexaScribe on this page, because this page measures price and nothing else. No affiliate relationships with any vendor listed; received no compensation for inclusion, ranking, or placement. Outbound vendor links use rel="noopener" only (not nofollow) — competitors are authoritative entities and there's no SEO reason to withhold authority signals. Editorial standards: see our editorial standards.

Revision history. Published May 14, 2026 with prices verified that week. Market shifts recorded at the time: Rev moved from per-minute PAYG to a subscription-only AI tier; Deepgram Nova-3 launched at $0.0077/min batch; AWS Transcribe Tier 1 settled at $0.024/min. Updated October 4, 2026 — structural and accuracy corrections only, detailed above, with no price re-verification. The badge at the top of the page carries both dates separately so they cannot drift apart again.

Where to go next

Best affordable transcription software

You know the prices — this ranks the tools behind them, and what you give up at each tier

VexaScribe pricing

All four tiers in full, plus the team plan the tables above don't cover

AI vs human transcription

The quality argument this page only costed — 10 scenarios, decision framework

Best transcription APIs

If the PAYG rows won on your volume, 12 APIs compared on more than price

Whisper transcription

What you're actually running if you self-host, before you price the GPU

Free transcription

The free paths in full, with what each one costs you instead of money

13 best transcription tools, 2026

Otter, Rev, Descript, Granola, AssemblyAI, Deepgram and others, reviewed

Editorial standards

How we verify prices, what we refuse to publish, and how to report an error

Também disponível em português — Quanto custa a transcrição — preços por minuto, hora e plano