Per character, per month, or per hour: how AI narration is priced

Shop around for AI narration and you meet three price tags that don’t line up. Cloud platforms charge by the million characters. Voice apps sell a monthly subscription with a usage allowance inside it. A few services, ours included, charge by the finished hour of audio. All three do the same thing: a machine reads text aloud. But run one ordinary novel through each and the totals differ by fifty times. Here is why.

The industry charges by the character

Most of the market charges for the input: you pay for the text you put in, character by character. As of mid-2026, the big cloud providers charge about $15 to $30 per million characters for their current voices. Older budget voices run around $4, and premium studio tiers reach $160. The specialist voice platforms cost more: pay-as-you-go sits near $100 per million characters, and their subscriptions work out to $150 or more per million once you divide the monthly fee by the allowance.

Per-character pricing isn’t a trick. For a developer buying raw synthesis one API call at a time, it’s the natural unit. The trouble starts when the customer is a reader, not a developer, and what they bring is a whole book.

How big a book is, in characters

A typical novel runs 90,000 to 120,000 words. That’s roughly 500,000 to 750,000 characters. Put that through the meter and the totals jump out. At big-cloud rates, voicing one novel once costs $9 to $18. At specialist-platform rates, $60 to $100. Services that turn a manuscript into an audiobook for an author charge anywhere from about $40 to a few hundred dollars per book.

Subscription allowances are the same math, just less obvious. A mid-tier creator plan at around $22 a month usually includes about 120,000 characters. That sounds generous until you notice it’s about a fifth of one novel. Someone who just wants to hear a book they own would spend five months of allowance on it. (The audiobook-subscription world has the same trap, paying every month for hours you may never use. We break it down here.)

None of this is hidden. It’s just a pricing model built for thirty-second clips, ads, and app voices, now being handed an eighty-thousand-word book.

Convert it all to listening time

What a reader actually cares about is the hour of listening. At ordinary narration speed, one hour of finished audio is roughly 55,000 characters of English. Convert the per-character prices and you get about $0.85 to $1.70 per audio-hour at big-cloud rates, and $5 to $9 per audio-hour on the specialist platforms. A ten-hour novel, one voice, no separate character casting, lands between $9 and $90, depending on whose meter you used.

BookNPC charges 25 cents per finished audio-hour for the same thing, metered to the second. That ten-hour novel is $2.50, paid once, and the audio is yours to keep and replay offline. Giving every character their own voice is a separate one-time analysis, about 3 cents a page, also paid once. In the app that novel runs to about 130 pages, so casting it comes to about four dollars, and fully cast and rendered in studio audio it is under seven dollars all in.

Why the gap is real, not a promotion

Two technical choices create it.

The first is what we meter. Per-character pricing charges for the input text, whatever it costs to synthesize. We charge for the output, the seconds of finished audio you actually get. We can, because we run open-weight speech models on our own GPUs, and on modern hardware the compute behind an hour of audio is small and still falling. Open models have caught up with the proprietary ones on quality. Once that happened, the expensive part of the old setup was no longer the synthesis. It was the pricing model wrapped around it. Billing by finished audio passes that efficiency on to you instead of keeping it.

The second is where the hard thinking happens. Making a full-cast audiobook is really two jobs. The first is working out who speaks each line of the novel. That’s a reading-comprehension problem, so we run it once as a language-model analysis, priced per page and paid once per book. The second is turning text into a voice. That’s a rendering problem, and it’s priced per hour only if you choose the studio render. Once a book is cast, it plays with free on-device voices forever. The cloud render is an optional upgrade, not a toll gate.

And the plain narrator is free, with no hour caps, because on-device speech costs nothing to run. Your phone already ships with capable neural voices, and no server is involved. Charging a subscription for that would just be charging for the meter.

The honest caveat

If what you need is short clips, not books, BookNPC is the wrong tool. But you still don’t have to feed a per-character meter. Our companion app VoiceNPC runs the same speech model entirely on-device (iPhone and Apple Silicon Macs), can clone a voice, and its free tier covers that kind of short work. The mismatch in this post shows up only when the text is book-sized, which is the only size BookNPC cares about. The numbers above were read off public price pages in July 2026 and rounded. Prices move, so check the current ones before you rely on any single figure. Ours are on the pricing page, and the free on-device narrator is the same price it always has been: nothing.