Gemini 3.8 Flash TTS API: Pricing and Setup

Gemini 3.8 Flash TTS is Google's flagship Gemini 3.8 speech model, published 23 September 2026. The API id is gemini-3.8-flash-tts. Standard paid pricing through 31 December 2026 is $0.50 per 1M text input tokens and $9.00 per 1M audio output tokens.

Gemini 3.8 Flash TTS is not on RouterPlex yet. Register for the current 60+ models. The live chat model Gemini 3.8 Flash is gemini-3.8-flash at $0.75 / $3.75 and returns text. The speech id can be used here once the marketplace lists it. Prepaid, 0% markup on the vendor list price, $0 top-up fee, hard spend limit.

Gemini 3.8 Flash TTS API pricing #

Paid rates, USD per 1M tokens. Audio tokens are 25 tokens per second. Google prints the 10-second equivalent next to the output price: $9.00 × 250 / 1,000,000 = $0.00225.

Storage on the caching rows is $0.50 per 1M tokens per hour through 31 December 2026, then $1.00. Flex cache input is cheaper than batch cache: $0.025 then $0.05, versus batch cache at $0.0625 then $0.125. Priority cache input is $0.225 then $0.45. Priority's 10-second equivalent is $0.00405 through 31 December 2026, then $0.0081.

Google's standard free tier lists input and output as free of charge, and marks Used to improve our products: Yes. The paid column marks that flag No. Batch free tier is not available. That free tier is Google's. It is not a RouterPlex credit.

What the model is #

Older Gemini TTS models returned headerless PCM by default. If your code wraps raw PCM in a WAV header, Google says to stop wrapping and write the bytes, or set the response format explicitly if you still need headerless PCM.

Both 3.8 TTS models share a schema with Flash-Lite TTS. Put sustained style and the speaker name in speech_metadata. Leave only point events in the transcript. A laugh, a sigh, and a short pause are the examples Google gives. A line such as "Say cheerfully: Hello" can be spoken aloud.

The 23 September post also describes 2,000+ production voices, including Mexican Spanish, Quebec French, and Scots English, plus voice replication from a 30-second sample with consent verification, SynthID, and C2PA. Every Gemini Audio clip is watermarked with SynthID. The model card is Gemini 3.8 Audio. Voice remixing was marked coming soon on that post.

Hume Voice Design chart #

Google's chart, methodology at deepmind.google/models/evals-methodology/gemini-3-8-tts:

ElevenLabs leads the Voice Qualities row. Google leads the other three rows on this chart. The blog also says Flash and Flash-Lite placed first and second on Hume's Overall Quality Index. That index is not in the table above, and Google did not print its scores.

Character pricing for ElevenLabs is on the Eleven v4 page. Different unit, different bill.

Call the live chat model #

gemini-3.8-flash at $0.75 / $3.75 is the chat route. Gemini 3.8 Live is a different product and is already covered.

curl https://api.routerplex.com/v1/chat/completions \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.8-flash",
    "messages": [
      {"role": "user", "content": "Write a two-sentence studio prompt for a narrator voice."}
    ]
  }'

On RouterPlex, once Flash TTS is listed #

The speech route gets a name when the row exists. It will not be silently aliased to gemini-3.8-flash. Billing will be the vendor list price, 0% markup, prepaid, hard per-key budget. Create an account for the 60+ models live now. The catalog and the OpenAI-compatible API are the contract those rows already use.