Fish Audio's Live 3 Builds a Voice AI Agent in 30 Seconds

Fish Audio has launched Fish Agent, a public-beta, no-code platform for building AI voice agents that place and answer phone calls. It runs on Live 3, the company’s new full-duplex speech model for real-time conversation. Fish says developers can describe an agent, assign it a phone number and publish it in about 30 seconds.

Live 3 targets a persistent problem in voice applications: deciding when to speak. Fish reports that the model can choose a brief reaction such as “mm-hmm” or “okay, 415” in as little as 186 milliseconds. A full-duplex model continues processing incoming speech while generating audio, allowing it to acknowledge, wait or stop as a conversation unfolds.

Turn-taking moves into the model

Conventional voice agents often chain automatic speech recognition, a large language model and text-to-speech. A voice activity detector watches volume and silence to estimate when the caller has finished. Brief thinking pauses can trigger premature replies, while short acknowledgements such as “uh-huh” can be mistaken for a new turn.

Live 3 puts those floor-control decisions inside a speech model trained on human conversations. According to Fish, it can listen while speaking and choose among acknowledging the caller, continuing its response, yielding the floor or waiting.

From prompt to phone line

Fish Agent packages Live 3 with the components needed to deploy a working application. Developers configure a system prompt, select or clone a voice, attach reference material, connect tools and publish the result to the web, a phone number or an application.

Connecting telephony over REST

Phone-number assignment uses the REST endpoint at /v1/agent/phone-numbers. The following request connects a Twilio number to an existing agent:

curl -X POST https://api.fish.audio/v1/agent/phone-numbers \
  -H "Authorization: Bearer $FISH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "twilio",
    "phone_number": "+14155550123",
    "agent_id": "agt_a1b2c3d4e5f6"
  }'

Launch pricing under four cents

Fish lists the combined speech-recognition, LLM and voice service at $0.035 per minute during its launch promotion. The standard rate is $0.07 per minute, with telephony and number rental billed separately.

The API is hosted at api.fish.audio under /v1/agent/*. Fish has stated the promotional rate ends December 31, 2026.

The 186 ms claim needs context

The reported 186 ms measures the model’s decision to produce a short reactive utterance. The announcement does not provide time to first audio for a complete answer, tail latency under load or interruption accuracy. Those measurements have a direct effect on whether an agent talks over callers or leaves long gaps.

Fish has also published no benchmark scores for Live 3. At launch, Artificial Analysis has no Live 3 entry in its duplex-speech harness, leaving the available performance figures vendor-reported. A production evaluation should measure:

• Median and 95th-percentile time to first audio for complete responses.
• False interruptions during pauses, background noise and brief acknowledgements.
• How quickly and reliably the agent stops when the caller takes the floor.
• Tool-call accuracy, task completion and recovery from API failures.
• Audio quality over real phone networks, including noisy and low-bandwidth calls.

One stack, one runtime

Fish Agent competes with OpenAI’s Realtime API, ElevenLabs’ Realtime v1, Google’s Gemini Live and agent platforms built around LiveKit or Vapi. Fish supplies the duplex model, agent runtime, telephony, testing and observability as one hosted service. It also advertises an on-premises version of the integrated stack for enterprise customers.

Live 3 has no standalone real-time API at launch. Access runs through Fish Agent, so developers choosing the model also adopt Fish’s runtime and deployment model. A modular architecture can still combine separate speech recognition and LLM services with Fish Audio text-to-speech, or use Fish components inside another agent framework.

Production buyers will also need to confirm data-retention policies, deployment regions, compliance certifications, concurrency limits, rate limits and service-level commitments. The public launch details do not specify those operational terms.

Where the beta fits

Fish Agent’s initial feature set fits voice support, outbound sales, scheduling and reminder systems that operate in English. Its knowledge grounding, tool calls, telephony and pre-release tests cover the main components of a transactional voice workflow without requiring separate vendors for each stage.

Applications requiring multilingual conversation cannot use Live 3 for that requirement at launch, despite the broader language coverage in Fish Audio’s voice catalog. Teams that require independently verified latency or quality scores also lack enough public evidence for a benchmark-based vendor decision.

Early pilots should compare Live 3 with the existing production stack using recorded call scenarios, noisy phone audio, caller interruptions and real tool failures. Those tests will show whether its learned turn-taking improves completed tasks and call flow beyond Fish Audio’s controlled demonstrations.