Gemini 3.8 Flash TTS free tier vs paid: test five lines, then budget

Yes: Google's pricing page (read 2026-10-09) lists Gemini 3.8 Flash TTS and Flash-Lite TTS as "Free of charge" for both input and output on the free tier, and bills the paid tier at $0.50 per million text tokens in and $9.00 (Flash) or $6.00 (Lite) per million audio tokens out through December 31, 2026. Sume has no free tier for TTS in its catalog: each job is billed, with a minimum of 1 cent.

That makes the free tier a fine place to try a voice and a poor place to plan a production run, because a free tier is a different service level from the paid rate. This post lays out what to test for free and what the paid line looks like, then what the same test costs on Sume.

What the pricing page says

For Gemini 3.8 Flash TTS the page lists a Standard tier with a free column and a paid column. The paid column reads $0.50 per million text input tokens through December 31, 2026 and $1.00 starting January 1, 2027. Audio output is $9.00 per million audio tokens, then $18.00. The page states the same as $0.00225 per 10 seconds of audio, then $0.0045.

Flash-Lite TTS has the same input rate, with audio output at $6.00 then $12.00 per million, or $0.0015 then $0.003 per 10 seconds. Both models count audio at 25 tokens per second.

A five-line test on each

Take five test lines, each 200 characters, and say each runs about 12 seconds (your own assumption; measure it). That is 60 seconds of audio, or 1,500 audio tokens. On the paid Flash rate that is 1,500 / 1,000,000 x $9.00 = $0.0135 in audio output; on the free tier it is not billed.

On Sume the same five lines are 5 x 200 = 1,000 characters. At $0.0475 per 1,000 characters (Sume API catalog, read 2026-10-09), the arithmetic is $0.0475, but the catalog lists a 1-cent minimum per job, so five separate jobs cost at least 5 cents by that floor. A single job with all five lines in one transcript is 1,000 characters.

How to use the free tier well

Use it to choose a voice and check pronunciation of your brand names. Do not treat it as a plan for production volume, since the page lists it as a separate tier from the paid one. Once the voice is chosen, move the real run to a paid key and budget with the table above.

If you need speech inside a pipeline that also cuts video and burns captions, Sume's TTS returns word timings on request (timestamps.words) and sentence slices as wav, which the video captions job can use. That is a workflow choice, not a price claim.