Today’s editionWed 11 Nov 2026
AI Compass
WeeklyArchiveAbout

Company dossier

Fish Audio

AI Compass tracks 1 confirmed launch from Fish Audio in model (11 Nov 2026), verified against primary official documentation and release notes.

1Launch

1Category

Official domain
fish.audio
Latest coverage
11 Nov 2026
Focus areas
Models 1

Peer labs in Models: OpenAI (38) · Anthropic (35) · Google (24) · NVIDIA (10) · Meta (9) · Google DeepMind (8)

Fish Audio launches full-duplex Live 3 voice model API with 186 ms response decisions

Published on 10 November 2026 for English voice agents on the Fish Agent Platform and API, Live 3 trains conversational turn-taking directly into the speech model so agents listen while speaking.

ModelStory 6 of the day

60-second takeaway: Fish Audio releases its full-duplex Live 3 English voice model API, deciding when to acknowledge, wait or yield in as little as 186 ms.

Why it matters: Learning backchannels, hesitation pauses, and true barge-ins inside the speech model removes brittle silence-timeout tuning that causes support and ordering voice bots to talk over callers.

Architecture and language
English full-duplex conversational speech model paired with a separate agent reasoning and tools layer
Turn-taking latency
Learned Acknowledge, Wait and Yield decisions in as little as 186 ms across 2 million+ voices
Platform promo pricing
$0.035 per minute on Fish Agent Platform (50% off $0.07 standard rate through 31 December 2026; $5 free credit)
Telephony add-ons
$0.01 per minute for PSTN calls; $1.20 per month rented Twilio numbers or free bring-your-own-carrier SIP