Cost of Transcription Per Hour in 2026: Real Numbers

An hour of audio transcription costs anywhere from $0.21 to over $119 depending on who does the work. The three-order-of-magnitude spread is not hype; it reflects three fundamentally different cost structures: raw GPU compute, packaged consumer software, and human labor. This post walks through the math for each so you can calculate your actual per-hour rate.

Step 1: Know Which Cost Model You Are Buying

Before comparing numbers, identify which pricing model applies to you.

Metered (per-minute APIs): You pay for exactly what you use, billed per second or per minute. Cost scales linearly with volume. Best for developers, low-volume users, and anyone building a product.

Flat-rate subscriptions: A fixed monthly fee for unlimited or capped minutes. Effective per-hour cost falls as you use more. Best for sustained high-volume users.

Human transcription: Pay per audio minute for a person to transcribe. Cost is mostly fixed by labor; no volume discount meaningful enough to approach AI rates.

Self-hosting models on your own GPU infrastructure is a fourth option covered at the end.

Step 2: Look Up Your Actual Per-Minute Rate

The table below shows verified rates as of July 2026 for pre-recorded (batch) transcription at standard pay-as-you-go tiers.

Notes: AssemblyAI plans a 10% in-region price increase effective July 1, 2026. AWS pricing uses US East (N. Virginia) tier-1 rate; volume tiers drop this to $0.0078/min at 5M+ minutes per month. Google Cloud Dynamic Batch drops to $0.004/min for non-time-sensitive workloads with a 24-hour turnaround. Deepgram Nova-3 streaming is cheaper ($0.0048/min) but charges for connection time, not just audio duration.

For a broader cross-vendor view of API rates, see the speech-to-text API pricing breakdown for 2026.

AssemblyAI U-3.5

$0.21/hr

OpenAI Whisper

$0.36/hr

Deepgram Nova-3

$0.46/hr

Google Cloud STT

$0.96/hr

Rev AI

$15/hr

Rev Human

$119/hr

Metered API rates at standard pay-as-you-go. Human transcription at Rev standard rate. Self-host estimate at cloud GPU utilization.

Step 3: Calculate Your Effective Rate for Subscription Plans

Metered APIs look cheapest on paper. But if you use a flat-rate subscription, your real per-hour cost depends on volume.

The formula: monthly plan cost / hours used per month = effective hourly rate.

TurboScribe's unlimited plan costs $10/month (billed annually). Otter Pro costs $8.33/month (billed annually, 1,200 minutes included). These have different effective rates depending on usage:

*Otter Pro includes 1,200 minutes (20 hours) per month; usage above that requires a higher plan.

The subscription becomes cheaper than Deepgram Nova-3 PAYG ($0.46/hr) somewhere around 22 hours per month for TurboScribe. It beats OpenAI Whisper ($0.36/hr) around 28 hours per month.

For a deeper look at when flat-rate wins versus metered, see unlimited vs. metered transcription pricing.

Step 4: Map Your Usage Profile to the Right Model

Three real-world scenarios with the math worked out.

Scenario A: Solo podcaster, one 60-minute episode per week Monthly volume: ~4 hours.

• AssemblyAI PAYG: 240 min x $0.0035 = $0.84
• OpenAI Whisper API: 240 min x $0.006 = $1.44
• TurboScribe Unlimited: $10.00
• Otter Pro: $8.33

At this volume, a metered API costs under $1.50 per month. A subscription adds features (AI notes, integrations, editing) that may justify the difference, but the raw transcription cost favors APIs by a wide margin.

Scenario B: Journalist with 20 hours of interviews per month Monthly volume: ~20 hours (1,200 minutes).

• AssemblyAI PAYG: 1,200 min x $0.0035 = $4.20
• OpenAI Whisper API: 1,200 min x $0.006 = $7.20
• TurboScribe Unlimited: $10.00

Costs are now close. Subscriptions add predictable billing and extras; APIs remain cheaper on raw cost. Decision factors: do you want variable or fixed bills, and does the platform's feature set matter?

Scenario C: Content team transcribing 60 hours of footage monthly Monthly volume: ~60 hours (3,600 minutes).

• AssemblyAI PAYG: 3,600 min x $0.0035 = $12.60
• OpenAI Whisper API: 3,600 min x $0.006 = $21.60
• TurboScribe Unlimited: $10.00

A $10/month unlimited subscription now costs less than even the cheapest API. The effective rate on TurboScribe is $0.17/hr, beating every metered option. This is where flat-rate plans decisively win.

The hidden costs of transcription services guide covers what else to watch for: export format fees, storage limits, and seat-based pricing that can quietly double the apparent per-hour rate.

When Per-Minute Pricing Actually Wins

Metered APIs make sense in three specific situations.

First: very low or unpredictable volume. Under about five hours per month, a PAYG API costs less than almost any subscription. A consultant who has three busy months and nine empty months should not pay subscription fees for nine months of zero use.

Second: building a product on top. If you are embedding transcription into your own software and passing costs to customers, per-minute pricing aligns with your unit economics. A flat subscription means eating the cost when a single heavy user spikes volume.

Third: experimenting with multiple providers. APIs let you benchmark AssemblyAI against Deepgram against Google Cloud on the same audio without committing to a subscription. The best speech-to-text APIs in 2026 has side-by-side accuracy comparisons.

Why Human Transcription Costs What It Costs

A trained transcriptionist takes three to six hours to produce one finished hour of transcript. At modest wages, the labor alone is $40 to $80 per audio hour. Add quality review, project management, and margin, and Rev's $1.99/min ($119.40/hr) is not surprising.

The human premium is justified in a narrow set of cases: legal depositions where errors have formal consequences, medical documentation with clinical stakes, broadcast captioning under regulatory standards, and heavily accented speech on degraded recordings that AI still handles poorly. For most clean audio, AI accuracy is high enough that paying 250x the API rate is hard to justify.

For a rigorous breakdown of when accuracy requirements actually call for human review, see AI vs. human transcription.

The Self-Host Option

Running Whisper Large-v3 on a rented GPU (a Spheron L40S, for example) costs roughly $0.02 per audio hour at full utilization. That is 10x cheaper than the cheapest commercial API. The catch: you need engineering capacity to set up infrastructure, monitor GPU instances, handle edge-case audio, and keep model weights current. The DevOps overhead typically costs more than the compute savings unless you are already running GPU workloads or transcribing well above 2,000 hours per month.

Most organizations that evaluate self-hosting either commit fully (building an internal transcription service) or stick with managed APIs for everything.

A Starting Point

If you just need a clean transcript without managing API keys or metered billing, ConvertAudioToText offers a browser-based tool with a free monthly allowance. It works well for one-off transcription without a subscription.

For regular high-volume work, the math in this post gives you the break-even points. Run your actual monthly hours through the formula, compare to your top two or three candidates, and pick the cheapest option for your usage profile. The transcription pricing comparison for 2026 has the full cross-vendor table if you want to go deeper.

Frequently Asked Questions

How much does transcription cost per hour in 2026?

The range is wide: developer APIs start around $0.21 to $0.46 per hour for pre-recorded audio, consumer AI tools like Rev charge around $15 per hour at pay-as-you-go rates, and human transcription runs $90 to $120 per hour. Unlimited subscription plans can push your effective per-hour cost below $0.20 if you transcribe enough volume monthly.

Is it cheaper to use a developer API or a flat-rate subscription?

It depends on volume. If you transcribe fewer than five to eight hours per month, a metered API is almost always cheaper than a $10 to $30 monthly subscription. Above ten hours per month, most subscriptions become competitive. Above 30 hours, an unlimited subscription typically wins on price.

Why is human transcription so much more expensive than AI?

A human transcriptionist takes three to six hours to produce one hour of finished transcript. Even at modest wages, the labor cost alone reaches $40 to $80 per audio hour. Company overhead, quality review, and margin push the consumer price to $90 to $120 per hour. AI transcription substitutes GPU compute for labor, dropping the cost by two orders of magnitude.

What is the cheapest AI transcription API per hour?

As of July 2026, AssemblyAI's Universal-3.5 Pro async model costs $0.0035 per minute ($0.21 per hour). Deepgram Nova-3 pre-recorded costs $0.0077 per minute ($0.46 per hour) at pay-as-you-go. OpenAI whisper-1 and gpt-4o-transcribe are each $0.006 per minute ($0.36 per hour). Prices may shift with volume tiers and growth plan discounts.

When does it make sense to pay for human transcription?

Human transcription is justified in a narrow set of cases: legal depositions or court filings where errors have formal consequences, medical documentation where clinical accuracy is critical, broadcast captioning under regulatory standards, or heavily accented speech on degraded audio that AI models still stumble on. For clean audio in most other contexts, AI accuracy is high enough that the human premium is hard to justify.

How do I calculate my true per-hour transcription cost?

For metered plans: multiply your per-minute rate by 60. For flat-rate plans: divide your monthly bill by your actual hours used. If you transcribe 10 hours on a $10/month plan, your effective rate is $1/hr; at 50 hours it drops to $0.20/hr. The key variable is your actual usage, not the plan's nominal rate.

Sources

• Deepgram pricing page: https://deepgram.com/pricing
• AssemblyAI pricing page: https://www.assemblyai.com/pricing
• OpenAI Whisper API pricing (brasstranscripts.com): https://brasstranscripts.com/blog/openai-whisper-api-pricing-2025-self-hosted-vs-managed
• AWS Transcribe pricing page: https://aws.amazon.com/transcribe/pricing/
• Google Cloud Speech-to-Text pricing: https://cloud.google.com/speech-to-text/pricing
• Rev pricing page: https://www.rev.com/pricing
• TurboScribe pricing: https://turboscribe.ai/pricing
• Otter.ai pricing: https://otter.ai/pricing
• Happy Scribe pricing: https://www.happyscribe.com/pricing