Gemini 3.8 TTS: design a voice from a prompt, live in the API
Two new Flash models clone a voice from 30 seconds and speak 100+ languages — plus AWS's cheaper Claude Code rival and Ollama's new thinking-level API.
Copy markdown
Two models, two jobs
Gemini 3.8 Flash TTS handles creative direction and character voices; Flash-Lite TTS is tuned for high-volume, cost-efficient scale. Both are live now in the Gemini API and AI Studio, and every clip ships with a SynthID watermark.
Design a voice from a sentence
Describe a voice in plain language and it builds one, pick from 2,000+ ready-made voices, or clone a voice from a 30-second sample with consent verification. Line-by-line control of pacing and emotion holds across hours of audio, and it took #1 on Hume AI's Voice Design Benchmark at 71.4.
Build this weekend
One API now covers dubbing, audiobook narration, and in-app character voices across 100+ languages, with no separate TTS vendor to stitch in. Voice cloning plus per-line direction turns a prototype narrator or a localized voiceover tool into a weekend project instead of a pipeline.
Elsewhere: AWS undercuts Claude Code
AWS open-sourced Strands Harness (Apache-2.0), a bring-your-own-model agent that runs on Bedrock, Anthropic, OpenAI, Google, or Ollama from one line of setup. The team reports ~28% lower token cost at comparable accuracy, and up to 77% cheaper than Claude Code on Terminal Bench 2.1 with the Fable 5 model. Install it with `pip install strands-harness`.
Elsewhere: Ollama exposes thinking levels
Ollama v0.34.3 now advertises each model's thinking levels and defaults through `GET /api/show`, so your app can read them at runtime instead of hard-coding, and it adds Nemotron-H vision on Apple Silicon via MLX.