Google ships 2 Gemini 3.8 speech models, says they top Hume benchmark
Both models are live in the Gemini API and Google AI Studio, and Artificial Analysis puts one of them first on its pronunciation benchmark.
Google released two text-to-speech models on Wednesday, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. Both are live in the Gemini API and Google AI Studio. The company says they cover more than 100 languages and dialects, among them Quebec French and Scots English.
Artificial Analysis, which tests models independently, said Gemini 3.8 Flash TTS debuted first on its benchmark for how reliably a model pronounces hard text. It came second on the provider voice arena leaderboard. On a controlled voice arena, Gemini 3.8 Flash-Lite TTS took first in Japanese at an Elo of 1,212, with the larger model second.
Google says a user can write a prompt to create a voice, then direct the delivery line by line. It says a voice can be replicated from a 30-second sample after what it describes as consent verification. The library runs from 30 original voices to more than 2,000 production-ready ones, and every clip carries a SynthID watermark, according to the company.
Logan Kilpatrick, who leads Google AI Studio, said on X that the models cost less than the earlier Gemini 3.1 Flash TTS. Google's price list puts audio output for Gemini 3.8 Flash TTS at $9.00 per million tokens through 31 December 2026. Gemini 3.1 Flash TTS is listed at $20.00. The new price rises to $18.00 in January.
Artificial Analysis priced the models per million characters of input text rather than per token, and reached the opposite comparison. It put Gemini 3.8 Flash TTS at $32.98 and the Flash-Lite model at $22.07, against $18.31 for Gemini 3.1 Flash TTS. It measured the larger model at 44.1 characters a second, behind Falcon 2 at 204.9.