Google's Gemini 3.8 TTS models with two-speaker dialogue arrive on Vercel's AI Gateway

Vercel has added two of Google's speech models to its AI Gateway: Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS. According to the Vercel blog, both take text as input and generate audio in more than 100 languages, with support for long-form narration, control over delivery, and two-speaker dialogue.

Two variants, two jobs

Per the changelog, the split between the models comes down to workload. The lighter option, google/gemini-3.8-flash-lite-tts, is aimed at high-volume speech generation and exposes controls for tone, pacing, and line-by-line delivery. That profile suits pipelines that regularly render large batches of text into audio and need consistent, tunable output.

The larger model, google/gemini-3.8-flash-tts, leans into expressiveness. It supports voice and character design driven by natural-language prompts, so instead of choosing from fixed presets, developers can describe acting cues, accents, and conversational reactions — effectively directing a performance in prose.

Both models share the headline capability of two-speaker dialogue, meaning a single generation can render a back-and-forth exchange between two voices rather than a lone narrator. Combined with multilingual output and long-form narration support, that covers the core requirements of audiobook production, podcast tooling, dubbing workflows, and voice agents.

Getting started

Vercel's documented example calls the experimental_generateSpeech function from its AI SDK, passing the gateway model identifier, the text to speak, a voice name — the sample uses "Kore" — and an output format such as WAV. The call returns audio data that can be written directly to a file.

For teams evaluating the models before writing code, Vercel provides a playground for each variant where speech can be generated and previewed, along with a speech quickstart and a dedicated text-to-speech guide covering setup and additional examples.

Gateway context

The release also fits into Vercel's broader positioning of AI Gateway as a single API for generation work across modalities. Speech requests run alongside a project's other models, with usage and cost tracked per request. Teams can configure routing rules to steer how requests are handled, and they can attach their own provider keys if they prefer not to rely on Vercel-managed access.

Why it matters

Speech is becoming a default output for AI products, from agents that respond by voice to pipelines that turn written content into audio editions. Google's Gemini 3.8 TTS models bundle multilingual coverage, two-speaker dialogue, and prompt-directed voice design, and Vercel's gateway makes them reachable through the same interface developers already use for other models. The practical benefit is consolidation: one integration, per-request cost visibility, and a choice between a bulk-oriented variant and an expressive one depending on the job at hand.