Eleven v3

Eleven v3 is the model to reach for when the reading matters as much as the words. Every other voice model here takes a delivery instruction as a separate field or as a preset: a style prompt, an emotion selector, a speed dial. This one takes it inline: write \`[whispers]\`, \`[excited]\`, \`[sighs]\` or \`[laughs]\` into the text and the tag steers the phrase it sits in front of. That is a finer instrument than a per-take mood, because a single line can turn twice.

The trade is that it wants more from you. A flat paragraph handed to v3 reads about as well as it does anywhere else; the model earns its price on text somebody has actually directed. It speaks every language this catalog offers, the only voice model here that does, across twenty-six preset voices, and it is the priciest per character apart from Gemini.

Reach for it on trailers, character work, anything with a turn in it. For long neutral narration where the point is clarity rather than performance, Gemini 3.1 Flash TTS is cheaper per minute of finished audio and needs no direction; for volume work, Kokoro 82M is a hundredth of the price. And if you need a voice that is not in the preset list, a clone or one of your own, MiniMax Speech 2.8 HD and Inworld TTS 2.0 are the two that accept a foreign voice id, which v3 on this path does not.