ElevenLabs v4 vs Gemini TTS: cost per hour with the assumptions shown

One hour of narration costs about $4.32 on ElevenLabs v4 at its listed $0.08 per 1,000 characters, assuming 54,000 characters per hour, and about $0.81 on Gemini Flash TTS output, assuming 25 audio tokens per second and $9.00 per million output tokens. Every number depends on assumptions you can change, so the script below shows them.

The rates come from Digital Applied's September tracker and the Gemini API changelog, both read 2026-10-02. The per-million-token unit for Gemini is our reading of the tracker, and the characters-per-hour figure is our assumption (about 150 words a minute at six characters per word).

What are the assumptions?

Change any row and the answer changes.

Why is the gap so large, and does it mean anything?

The two bills count different things: characters in, versus audio tokens out. Price per hour ignores voice quality, languages, voice control and limits, and the Gemini rates are promotional through 2026-12-31. Compare on a sample you listen to, not on a table.

chars_per_hour = 54_000  # assumption: 150 wpm x 6 chars
for name, per_1k in [('v4', 0.08), ('turbo', 0.04), ('v4 promo', 0.022)]:
    print(name, round(chars_per_hour / 1000 * per_1k, 2))
tokens_per_hour = 25 * 3600  # 25 audio tokens per second
for name, per_m in [('gemini flash', 9.00), ('gemini flash-lite', 6.00)]:
    print(name, round(tokens_per_hour / 1e6 * per_m, 2))

Where does Sume fit in a comparison like this?

Sume's TTS 1.0 lists a transcript limit of 20,000 characters per job and fails audio longer than 1,200 seconds with tts_duration_exceeded. The docs do not publish a Sume dollar rate for it, so there is no row here; run a short job and read the billed amount.