Transcription Pricing 2026: $0.15/hr APIs to $120/hr Human

Transcription pricing in 2026 spans three orders of magnitude. The same hour of audio can cost about $0.46 on a developer API or up to $120 on a human transcription service. The pricing model matters as much as the headline number, because per-minute pricing scales linearly with usage while flat-rate plans cap the cost. This post breaks down what each major service charges right now, how the structures work, and which model fits which use case. Every number here is checked against the vendor's own pricing page, with the source linked at the bottom.

The Short Version

There are only three ways transcription is priced, and the right one depends entirely on your monthly volume.

• Under a few hours a month: use a free tier or pay per minute. Developer APIs like Deepgram and OpenAI run about $0.36 to $0.46 per hour of audio.
• A few hours to a few hundred hours a month: a flat-rate consumer plan ($10 to $20 per month) is almost always cheapest, because per-minute pricing would pass the subscription fee.
• Legally or medically sensitive audio: pay for human transcription ($1.50 to $2.00 per minute, roughly $90 to $120 per hour) where 99 percent accuracy without manual review is the requirement.

My take in one sentence: if you transcribe more than three or four hours a month and you are not building a product on top of an API, a flat-rate unlimited plan removes the meter and the math stops mattering. I build ConvertAudioToText, so treat that as a disclosed bias, the comparison below names where rivals genuinely win.

The Three Pricing Structures

Every transcription service uses one of three pricing models.

Per-Minute (or Per-Hour) Pricing

You pay for each minute of audio you process. The unit cost is small (fractions of a cent per minute on the developer APIs) but the bill scales linearly with usage. Most enterprise and developer APIs work this way, and most bill per second so you are not rounded up to the nearest minute.

Best for: low-volume users who only need a few hours per month, and high-volume engineering teams with predictable usage and volume discounts.

Flat-Rate Unlimited

You pay a monthly subscription and process as much audio as you want. The effective unit cost falls as your usage rises, so the more you transcribe, the better the deal looks.

Best for: regular users who process five or more hours per month, where per-minute pricing would exceed the monthly subscription.

Hybrid Tiers

A free tier with a monthly cap (commonly 10 to 300 minutes), then a metered or flat-rate plan above that. This combines the on-ramp of free with the predictability of a subscription.

Best for: most consumers and small teams. Free for evaluation, fixed cost once they commit.

The Major AI Services Compared

A direct comparison of headline pricing for AI transcription in 2026. All per-hour figures are computed from the published per-minute rate at list price, before volume discounts.

AssemblyAI

$0.15/hr

OpenAI 4o-mini

$0.18/hr

AssemblyAI 3.5 Pro

$0.21/hr

OpenAI Whisper

$0.36/hr

Deepgram Nova-3

$0.46/hr

List price per hour, 2026. Lower is cheaper.

Two things stand out. The developer-API tier (Deepgram, the OpenAI models, AssemblyAI) is consistently the cheapest per-minute option, often under $0.50 per hour. The consumer-tier apps charge far more per minute on paper but bundle a UI, exports, editing, speaker labels, summaries, and integrations that a raw API does not include. A second, easy-to-miss point: the cheap API headline rate is the base rate. AssemblyAI's $0.15/hour becomes about $0.22/hour once you add speaker diarization, sentiment, and summarization, and Deepgram's mono rate is higher for multichannel audio. Compare like for like.

For deeper per-vendor breakdowns, our pages cover the details: TurboScribe alternatives, Otter AI alternatives, Happy Scribe alternatives, Rev alternatives, and Descript alternatives. The raw API rates are unpacked further in OpenAI Whisper API pricing and Deepgram Nova-3 explained.

Human Transcription Pricing

Human transcription remains the gold standard for accuracy in difficult audio. Pricing reflects the labor involved.

The "AI plus human review" tier produces near-perfect accuracy at a meaningful premium. For most use cases, modern AI accuracy on clean audio is high enough that the human premium is not justified. The exceptions are legal, medical, and accuracy-critical media work, where one missed word changes the meaning and a person has to stand behind the transcript.

Enterprise and Volume Pricing

For high-volume users (1,000-plus hours per month) the major API providers all offer custom pricing, and the discounts are real:

• Volume discounts of 20 to 50 percent off the headline per-minute rate. AWS Transcribe, for example, drops from $0.024/min to roughly $0.0078/min at the 5-million-minute tier.
• Reserved capacity pricing for predictable workloads.
• Annual commitments in exchange for further discounts.
• Deepgram's Growth plan already shaves the per-minute rate before any custom negotiation.

For enterprises at this scale, the choice is usually between bringing transcription in-house (running an open-source model like Whisper on owned GPUs) and using a managed service with volume discounts. Self-hosting Whisper can land near $0.006 per minute of compute, but only once you account for GPU time, engineering, and uptime, so the break-even depends heavily on operational maturity. Our post on enterprise transcription pricing covers this trade-off in detail.

What You Actually Get at Each Price Point

A pricing comparison is incomplete without comparing what each tier delivers.

Under $0.50 per hour (raw API tier)

Pure transcription, no UI, no exports beyond the raw API response, no built-in editor. Best for developers building their own products on top.

Examples: Deepgram Nova-3, AssemblyAI, OpenAI gpt-4o-transcribe.

$10 to $20 per month (consumer flat-rate)

Transcription with a UI, exports (SRT, VTT, TXT, DOCX), an editor, speaker labels, summaries, and integrations with common tools. Some tiers add AI templates tuned to use cases.

Examples: ConvertAudioToText ($9.99/mo yearly), TurboScribe ($10/mo yearly).

$25 to $50 per month (consumer premium)

The above plus advanced editing (Descript-style text-based editing), polished UI, more integrations, and sometimes light video editing.

Examples: Descript Creator and Business tiers, Happy Scribe Pro and Business.

$50 to $200 per month (business)

The above plus team features (shared workspaces, admin controls), higher usage caps, priority support, and SSO.

Examples: Otter Business, Descript Business, Happy Scribe Business, CATT Business.

$1.50 per minute and up (human transcription)

Hand-transcribed by trained humans, typically 99 percent accuracy or better. Slower turnaround, sometimes specialized for medical or legal work.

Examples: Rev Human, Happy Scribe Human, TranscribeMe, GoTranscript.

Cost Per Hour at Different Volumes

The break-even analysis is what most people actually need.

For most individual creators and small teams, a flat-rate unlimited plan ($9.99 to $20 per month) is the cheapest option the moment you pass roughly three or four hours a month. Our post on cost of transcription per hour walks through the math, and unlimited vs metered pricing covers when each wins.

On CATT pricing specifically, the Pro plan is $9.99/month billed yearly or $14.99 billed monthly for unlimited transcription, which covers anything from five hours to several hundred hours per month at a fixed cost. There is also a Starter plan ($5.99/mo yearly, 2 hours) for light users, a Developer plan ($15.99/mo yearly, 10 hours plus API keys), a Business plan ($47.99/mo yearly, unlimited plus webhooks and analytics), and a per-seat Team plan.

Hidden Costs to Watch For

Headline pricing does not capture the full cost picture. A few things can shift the actual cost meaningfully.

Add-on features. The cheap API headline rate is usually transcription only. Speaker diarization, sentiment, summaries, and translation are often billed on top. AssemblyAI's base $0.15/hour becomes about $0.22/hour with the common add-ons turned on.

Per-conversation caps. Otter's Pro plan caps a single recording at 90 minutes even though it allows 1,200 minutes a month, so a long lecture or all-day event can hit the ceiling on one file.

Storage and export fees. Some services charge to keep transcripts and audio long-term, or charge per export for premium formats. Most modern tools include all formats in the base price.

Overage fees. Tools that bill per minute sometimes add overage charges if you exceed an expected volume.

Multi-language surcharges. Some services charge more for non-English transcription. Most modern tools (including ConvertAudioToText) price all 99-plus languages the same.

Our post on hidden costs of transcription services covers these in detail.

Free Tier Comparison

Many services offer free tiers. The quality and the catch vary a lot.

TurboScribe's free tier is the most generous for casual use because 3 files a day at 30 minutes is roughly 45 hours a month if you use it daily, with no account barrier to start. CATT's free tier is 10 minutes monthly with no feature restrictions, plus a 30-minute preview you can run with no account at all. The free English transcription page is the standard entry point. For the trade-offs, see is free transcription worth it.

When to Pay vs Stay Free

The right answer depends on volume and feature needs.

Stay free if: you transcribe under 10 minutes a month, you mainly need a quick demo or a one-off transcript, and you can live with the feature set the free tier provides.

Pay flat-rate if: you transcribe more than a few hours a month, you need exports beyond plain text, you need reliable speaker labels, and you want predictable cost with no meter.

Pay per-minute if: you have variable usage that sometimes drops to zero, you are integrating transcription into your own product, or you need volume pricing at scale.

Pay for human if: you have legally sensitive transcripts, you have audio where AI accuracy is documented to fail, and you need 99 percent accuracy without manual review.

Our post on when to pay for transcription covers the decision logic.

Which Service Actually Wins for You

No single tool wins on every axis, and any comparison that puts the author's own product first should be read with suspicion. Here is the honest split:

• TurboScribe wins if you want unlimited transcription with an anonymous, no-signup free tier and the lowest annual price ($10/mo). It hands you the transcript and stops there.
• Otter wins if your core need is live meeting capture with a bot that joins Zoom, Google Meet, and Teams automatically. Its caps and per-seat pricing are the trade-off.
• Rev and Happy Scribe (human) win when you need certified, human-checked accuracy for legal, medical, or broadcast work, and you can absorb $90 to $120 per hour.
• Deepgram, AssemblyAI, and the OpenAI APIs win if you are an engineer building transcription into your own app and want the lowest per-minute cost.
• ConvertAudioToText is one of several solid flat-rate options. On annual billing it is roughly tied with TurboScribe ($9.99 vs $10), cheaper on monthly billing ($14.99 vs $20), and it adds AI templates, action items, a full editor, and first-class African-language support. If you only need a raw transcript and nothing else, a cheaper or free tool may serve you fine.

Where to Start

For most users, the cheapest fast evaluation path is the free tier of the flat-rate provider closest to your needs. Run a few representative files, see if the accuracy and feature set fit your audio, and commit to a paid plan only when you have evidence it will work for your use case. The 10-minute CATT free tier plus the 30-minute no-account preview is enough to test most common files. Our pricing page shows exactly what each paid tier delivers.

FAQ

How much does it cost to transcribe one hour of audio in 2026?

It ranges from about $0.36 to $0.46 per hour on a developer API (OpenAI, Deepgram), to roughly $12 to $15 per hour on consumer AI tools billed per minute, to $90 to $120 per hour for human transcription. On a flat-rate unlimited plan, the per-hour cost drops toward zero the more you transcribe, since you pay one monthly fee regardless of volume.

What is the cheapest way to transcribe audio?

For one-off or very light use, a free tier is cheapest (TurboScribe's 3 files a day or CATT's 10 minutes a month). For a few hours a month or more, a flat-rate unlimited plan around $10 to $20 per month is cheapest, because per-minute pricing would pass the subscription fee. For developers at high volume, a per-minute API like Deepgram at about $0.0077 per minute is the lowest unit cost.

Is per-minute or flat-rate pricing better?

It depends on volume. Per-minute pricing is cheaper when you transcribe only a little (under three or four hours a month) or when usage is unpredictable and sometimes zero. Flat-rate is cheaper and simpler once you pass that threshold, because the cost is capped no matter how much you process. The crossover for most consumer tools sits around three to four hours per month.

Why is human transcription so much more expensive than AI?

Human transcription is priced for labor: a person listens to and types out every word, then often reviews it. Rev charges from $1.50 per minute and Happy Scribe from $2.00 per minute, which works out to roughly $90 to $120 per hour. AI does the same job for cents because it runs on a model, not a person. You pay the human premium when 99 percent accuracy without manual review is a hard requirement, typically legal, medical, or broadcast work.

Do cheap transcription APIs have hidden costs?

Often, yes. The headline rate is usually for transcription only. Speaker diarization, sentiment analysis, summaries, and translation are frequently billed on top. AssemblyAI's $0.15 per hour base rate rises to about $0.22 per hour with diarization, sentiment, and summary enabled. Some providers also charge more for multichannel or multilingual audio, and AssemblyAI's in-region rates rise 10 percent on July 1, 2026 unless you request the global region. Always price the features you actually need, not the base rate.

How much does ConvertAudioToText cost?

CATT is free for 10 minutes of transcription per month (plus a 30-minute no-account preview). The Pro plan is $9.99 per month billed yearly, or $14.99 billed monthly, for unlimited transcription in 99-plus languages with speaker labels, AI templates, and all export formats. There are also Starter ($5.99/mo yearly, 2 hours), Developer ($15.99/mo yearly, 10 hours plus API), Business ($47.99/mo yearly, unlimited plus webhooks), and per-seat Team plans.

Is a free transcription tool good enough?

For short, clean, one-off audio, yes. Modern AI accuracy on clear English audio is high, and free tiers from CATT, TurboScribe, and Otter use the same underlying models as their paid tiers. The limits are volume (10 to 300 minutes a month, or per-file caps) and sometimes features. If you transcribe regularly or need exports, editing, and speaker labels at scale, a paid flat-rate plan pays for itself quickly. See our breakdown on is free transcription worth it.

Which transcription service is the most accurate?

For difficult audio (heavy accents, crosstalk, poor recordings) human transcription from Rev or Happy Scribe is still the most accurate at 99 percent or better. For clean audio, the gap has narrowed: top AI models (Deepgram Nova-3, AssemblyAI Universal, OpenAI gpt-4o-transcribe) reach the high 90s on clear speech, and the consumer tools built on them inherit that accuracy. Match the tool to your audio quality rather than chasing a single accuracy number.