Cheapest TTS API this week: Gemini Flash-Lite, $0.0015 per 10 seconds

Of the text to speech prices I read on 2026-10-05, Gemini 3.8 Flash-Lite TTS is the lowest per unit of audio: about $0.0015 per 10 seconds, which is $0.009 a minute and $0.54 an hour of output. Gemini 3.8 Flash TTS is about $0.00225 per 10 seconds, or $0.81 an hour. Google bills in tokens, not characters, so the per-10-second figures on its page are the comparable unit.

The prices

The page's per-10-second figures follow from the token prices at 250 output tokens per 10 seconds (25 tokens a second, my inference from its figures): 250 x $6.00 / 1M = $0.0015 and 250 x $9.00 / 1M = $0.00225. Per hour, 360 ten-second blocks x $0.0015 = $0.54 and 360 x $0.00225 = $0.81. These exclude text input tokens at $0.50 per 1M, a promo price through Dec 31, 2026.

Against Sume's Sonic route

Sume bills Cartesia Sonic per character: list $38 per 1M credits x 1.25 = $0.0475 per 1K characters, rounded up to the cent per job, plus a 5.5% platform fee. Cartesia states that a minute of audio takes 750-800 credits (one per character), so a minute is 0.75-0.8 x $0.0475 = $0.036 to $0.038 and an hour is about $2.14 to $2.28 (split into jobs of at most 20,000 characters). That is roughly four times the Flash-Lite hour of $0.54 (2.14 / 0.54 = 4.0, 2.28 / 0.54 = 4.2).

Sume does not ship Gemini TTS, so this is a price reference, not a Sume option. The extra cost buys Sume's managed pipeline, and Sonic is a different voice engine.

When the cheapest model is the right one

• Bulk narration where a neutral read is enough: long articles, internal training, accessibility audio.
• Drafts and animatics, where you replace the voice later.
• Not for hero ad reads until a listener has compared it with a stronger model on your scripts; a price of $0.009 a minute is irrelevant if the take is rejected.
• Always re-check the page: promo input pricing has an end date of Dec 31, 2026.