Gemini 3.8 Flash-Lite TTS: an audio hour is $0.54 now, $1.08 in 2027

On Google's Gemini API pricing page, Gemini 3.8 Flash-Lite TTS audio output is $6.00 per million tokens through December 31, 2026 and $12.00 from January 1, 2027. Google gives a per-second equivalent of about $0.0015 per 10 seconds in 2026, so one hour (360 ten-second blocks) of generated audio is about $0.54 in 2026 and $1.08 in 2027, before text input. The Flash tier is $0.81 and $1.62 per hour on the same arithmetic.

Prices below were read on Google's page on 2026-10-09. Text input is billed separately, at $0.50 per million tokens through 2026 and $1.00 from 2027.

The arithmetic

Google's page states $0.00225 per 10 seconds for Flash and $0.0015 per 10 seconds for Flash-Lite through the end of 2026, doubling afterward to $0.0045 and $0.003. One hour contains 360 blocks of ten seconds. So 360 x $0.0015 = $0.54, and 360 x $0.003 = $1.08 for Flash-Lite. For Flash, 360 x $0.00225 = $0.81 and 360 x $0.0045 = $1.62.

Why this does not map directly to Sume

Google prices audio output by the second of generated speech. Sume's TTS charges per transcript character, $0.0475 per 1,000 characters, and spaces and punctuation count. To compare the two for your own script you need the speech length of that script, which depends on voice, language and pace.

Because Sume does not list a Gemini speech model, this post makes no claim about its voice quality or about how Sume would compare for a given script. If you already know your script's character count and the audio length Google returned for it, the two prices can be set side by side: Sume's cost is characters / 1,000 x $0.0475.

What the price change means for planning

The doubling on January 1, 2027 matters most for work you can do now. Bulk narration generated in 2026 costs half of what the same output will cost in 2027, per Google's published rates. If you have a back catalog to voice, running it before year end locks in the lower rate, provided the quality fits your use.

• Budget 2027 at twice the 2026 audio-token price unless Google changes its page.
• Store generated audio so you do not pay again to regenerate it.
• Check the page again before a large run, since preview pricing can change.

Limits of this comparison

Everything above is audio-token pricing as Google published it for a preview period. It leaves out the text-input line, any free tier, retry costs, and the cost of checking the audio. It also assumes your generated speech really lasts an hour. A long script spoken at a fast pace will be shorter than the same script at a slow pace.

For a budget, run a 30-second sample, read the audio length Google returns, and extrapolate from that figure rather than from a words-per-minute rule.