Gemini 3.8 Flash-Lite TTS: 30,000 ten-second product clips cost $45

Thirty thousand ten-second product voice clips cost $45.00 in audio output on Gemini 3.8 Flash-Lite TTS through December 31, 2026, or $67.50 on Flash. From January 1, 2027 the same batch is $90.00 on Lite and $135.00 on Flash. Text input is billed separately on top.

Source: Google's pricing page (read 2026-10-09). Both models price audio at 25 tokens per second, so a 10-second clip is 250 tokens. Sume does not list these Gemini models.

Per-clip and batch arithmetic

One clip: 10 s x 25 tokens = 250 audio tokens. Flash-Lite at $6.00 per million: 250 / 1,000,000 x $6.00 = $0.0015, which matches the page's own note of $0.0015 per 10 seconds. Flash at $9.00 per million: $0.00225. After January 1, Lite is $12.00 per million ($0.003 per clip) and Flash is $18.00 ($0.0045).

Times 30,000 clips: $45.00 and $67.50 in 2026, $90.00 and $135.00 in 2027. The catalog is the unit here, not the product count: if half the products get a 15-second clip instead, the audio cost goes up by the extra 5 seconds x 25 = 125 tokens on those clips.

What a batch like this still needs

A synthesis price is only the first line. Clips of different lengths need trimming to the same slot, and a catalog of this size needs a way to find the one that failed. Keep a manifest that maps product id to the file you got back, and re-run only the missing ids.

If the finished clips go into video, each clip is just a file to a timeline. On Sume you import the audio and either concatenate up to 20 parts into one file with timeline audio at $0.01 a job, or place the parts inside a render with the timeline's audio.parts[]. That is $0.01 per 20 clips for the join, which is small next to the synthesis bill.

Checking the page before you commit

Prices on a preview model can change, and Google's own page labels the 2026 rates as running through December 31. Re-read the page the day you start the batch, and pin the numbers in your budget sheet with the read date, as this post does.

Budgeting the surprises

Real catalogs are not uniform. If 20 percent of the 30,000 products need a 20-second clip because the name is long, the audio volume is 24,000 x 10 s + 6,000 x 20 s = 360,000 seconds, which is 9,000,000 audio tokens, or $54.00 on Lite through December 31 (9 x $6.00). Compute the tokens from your real durations, not from the average you hoped for.

Also keep a retry reserve. A batch this size will have failures, and a 5 percent re-run is 1,500 more clips, which is $2.25 on Lite at the 2026 rate. Put that line in the budget before the run, not after.