Pricing and budget · 2026-09-22
Calculating the cost of a minute of narration on Gemini 2.5 Flash Preview TTS
Different from the Flash vs Pro comparison article, this focuses only on the Flash version and covers how Gemini 2.5 Flash Preview TTS prices input text and output audio separately, and how to estimate the per-minute cost of an in-app narration feature.
Two separate price components: input text and output audio
The model page states input text is billed at $0.50 per million tokens, and output audio at $10 per million tokens. Even though the output is 'audio,' the unit is still tokens; that means as the audio output's duration grows, its token count grows too, and cost scales accordingly. These two components need to be accounted for separately — estimating total cost by looking only at input text length misses the much higher unit price of output audio tokens.
In a usage pattern like in-app narration, a read-aloud feature, or an assistant voice flow, most of the total cost usually comes from output audio, since the output unit price is 20 times the input's.
How to estimate the per-minute cost
The exact token-to-minute ratio varies with speech rate and content; for a reliable estimate, generate a typical narration clip of yours first (a one-minute piece of text, for example), measure the output token count in the returned usage data, then multiply that figure by the output unit price. That measurement is a far more reliable base than budgeting on a general assumption.
Estimating a feature's daily or monthly total narration minutes and multiplying by that unit cost lets you forecast the feature's total operating cost upfront; that's a step worth taking before deciding whether to offer the feature free to users or behind a paid tier.
- Generate a typical clip and measure the returned output token count.
- Multiply that figure by the output unit price ($10 per million tokens).
- Most of the total cost comes from output audio, not input text.
function_calling support is an unexpected detail
The model page shows function-calling capability listed on this TTS model too; that can make it possible to build an integration that calls a tool before speaking and folds the returned result into the narration flow. Test the exact behavior and supported parameters in your own integration before going to production.
Frequently asked questions
Can I estimate cost by looking only at input text length?
No, output audio tokens' unit price is 20 times the input's; most of total cost comes from output. For a reliable estimate, measure the usage data returned from a real request.
How much cheaper is this model than the Pro version?
See the Flash vs Pro comparison article or the current pricing page for the exact price difference; this article focuses only on the Flash version's own cost calculation.