Gemini 3.8 Flash vs Flash-Lite TTS: $27 apart on 100 hours

For 100 hours of generated speech, Gemini 3.8 Flash TTS costs $81.00 in audio output at the 2026 price and Gemini 3.8 Flash-Lite TTS costs $54.00, a $27.00 gap. From 2027-01-01 Google's listed prices double, so the same 100 hours costs $162.00 and $108.00, a $54.00 gap. Text input is billed on top at the same rate for both models. Sume does not list either Gemini TTS model; this page is a budgeting aid for people choosing a voice model before they wire it into a video workflow.

Google's listed prices

Google's pricing page lists audio output per million tokens and also gives a per-10-seconds equivalent, which is what makes hour-level arithmetic possible without guessing a token rate.

The arithmetic for 100 hours

One hour is 3,600 seconds, so 100 hours is 360,000 seconds, or 36,000 blocks of 10 seconds. Multiply by the per-10-seconds figure.

• Flash, 2026: 36,000 x $0.00225 = $81.00.
• Flash-Lite, 2026: 36,000 x $0.0015 = $54.00. Flash-Lite is one third cheaper (6 / 9).
• Flash, 2027: 36,000 x $0.0045 = $162.00 (the page prices this at double).
• Flash-Lite, 2027: 36,000 x $0.003 = $108.00.
• The text you send in is extra: $0.50 per 1M input tokens in 2026, $1.00 in 2027. It is not included in the numbers above.

When the $27 matters

A $27 gap is small for a team that voices a few explainers a month. It matters for catalogs, language-learning libraries and per-listing ads, where hours of audio are generated every week. Because the discount ends on 2027-01-01, any batch you want at 2026 prices has to be generated and stored before then; stored audio is not repriced.

Price is one input. Google's page tells you nothing about how the two models sound, so run the same script through both and listen before you commit a library to the cheaper one.

How Sume prices the same hour

Sume bills text to speech by character: $0.0475 per 1,000 characters, with a 20,000 character limit per request (Sume API catalog, read 2026-10-09). The catalog lists TTS 1.0 and a TTS Router whose model ids are Sonic versions; it does not list a Gemini model. If you chose your own speech rate of 900 characters per minute, an hour is 54,000 characters, so 100 hours is 5,400,000 characters x $0.0475 per 1,000 = $256.50. That figure depends on your assumed rate, so replace 900 with a measured one from your own scripts.

If your pipeline needs the audio inside a Sume Timeline render, you can import an externally generated file and use it as the audio spine. See the Timeline audio docs for the join and split jobs, which cost $0.01 each.