Voice models

Phonic's end-to-end speech-to-speech model for agentic voice tasks, built to call tools while the caller is still speaking. It's served on Phonic's voice-agent platform.

Released

Oct 29, 2026

$ / min

$0.070/min

Big Bench Audio

98.9%

Liquid AI's small open-weight audio model. It takes speech or text and streams back text, speech or both, aimed at on-device voice chat.

Released

Oct 27, 2026

$ / min

—

Big Bench Audio

—

A full-duplex speech-to-speech model from Block's AI research lab, trained from scratch only on recorded human conversation, with no text and no pretrained models. It listens while it talks and runs locally on a high-end Mac.

Released

Oct 23, 2026

$ / min

—

Big Bench Audio

—

ElevenLabs' reasoning speech-to-speech model for voice agents. It bundles speech recognition, turn-taking, an LLM and voice in one model choice, and reasons before it answers.

Released

Oct 19, 2026

$ / min

$0.080/min

Big Bench Audio

98.8%

ElevenLabs' default, fastest and cheapest Eleven Realtime model, for assistants, tutors and interactive characters.

Released

Oct 19, 2026

$ / min

$0.080/min

Big Bench Audio

75.9%

An update to GPT-Live for ChatGPT, Codex and dots, with better accents, new voices and more natural delegation. API access is announced as coming soon.

Released

Oct 15, 2026

$ / min

—

Big Bench Audio

—

xAI's speech-to-speech model for voice agents, served through its Speech to Speech (Voice Agent) API. xAI also runs it on Starlink support lines and in Tesla cars.

Released

Oct 14, 2026

$ / min

$0.080/min

Big Bench Audio

98.5%

Amazon's latest speech-to-speech model on Bedrock, the successor to Nova 2 Sonic. It improves reasoning, instruction following, tool-calling accuracy and latency, with the same API, voices and languages.

Released

Oct 5, 2026

$ / min

—

Big Bench Audio

—

Alibaba's current full-duplex realtime voice model family on Model Studio, in Plus and Flash tiers. Alibaba points GPT-Live-1 and Gemini 3.8 Live users to the Plus tier.

Released

Sep 22, 2026

$ / min

—

Big Bench Audio

—

The realtime variant of Alibaba's Qwen3.8-Omni-Flash omnimodal model. It perceives and responds while receiving live audio and video streams, and can call tools along the way.

Released

Sep 18, 2026

$ / min

—

Big Bench Audio

—

Google's default Live API model for low-latency voice agents, built on Gemini 3 Pro. It takes live audio, images and video and answers in speech, and it also powers Search Live.

Released

Sep 15, 2026

$ / min

—

Big Bench Audio

91.7%

The higher-reasoning version of Gemini 3.8 Live, for complex multi-step voice tasks. It runs background reasoning at a configurable level while the conversation continues, and powers Gemini Live in the Gemini app.

Released

Sep 15, 2026

$ / min

—

Big Bench Audio

97.7%

Boson AI's audio-native speech-to-speech model and hosted API for voice agents, compatible with the OpenAI Realtime API.

Released

Aug 4, 2026

$ / min

—

Big Bench Audio

68.6%

NVIDIA's open-weight full-duplex speech model, an 11B hybrid Mamba/Transformer. NVIDIA calls it the first open full-duplex model that keeps the conversation going while it calls tools.

Released

Aug 3, 2026

$ / min

—

Big Bench Audio

27%

OpenAI's full-duplex voice model: it listens and speaks at the same time and hands deeper reasoning and tool use to a separate backend model. It powers ChatGPT Voice and has been in the API since September 2026.

Released

Jul 8, 2026

$ / min

$0.050/min

Big Bench Audio

90.1%

OpenAI's current Realtime API speech-to-speech reasoning model, with configurable reasoning effort and tool use. It updates GPT-Realtime-2 with better alphanumeric recognition, silence and noise handling, and interruption behavior.

Released

Jul 6, 2026

$ / min

—

Big Bench Audio

95.8%

A distilled, faster and cheaper version of GPT-Realtime-2.1 for realtime voice apps, with reasoning and tool use.

Released

Jul 6, 2026

$ / min

—

Big Bench Audio

74.7%

OpenAI's first Realtime API voice model with configurable reasoning, for speech-to-speech agents. It is still listed alongside its successor, GPT-Realtime-2.1.

Released

May 7, 2026

$ / min

—

Big Bench Audio

96.6%

Google's earlier audio-to-audio Live API model, still in preview. Google now labels it legacy and recommends Gemini 3.8 Live.

Released

Mar 26, 2026

$ / min

—

Big Bench Audio

96.6%

OpenAI's non-reasoning Realtime API speech-to-speech model for voice agents and customer support, still listed and not deprecated.

Released

Feb 23, 2026

$ / min

—

Big Bench Audio

81.4%

NVIDIA's open-weight 7B full-duplex speech-to-speech model, built on Kyutai's Moshi. A text prompt sets the persona and an audio prompt sets the voice.

Released

Jan 15, 2026

$ / min

—

Big Bench Audio

19.1%

Amazon's second-generation speech-to-speech model on Bedrock for real-time voice agents, with native telephony integrations. Nova 2.5 Sonic has since superseded it.

Released

Dec 2, 2025

$ / min

—

Big Bench Audio

88.1%

Kyutai's open-weight full-duplex spoken dialogue model, which models the user's and its own speech as parallel streams. It's the base that NVIDIA's PersonaPlex builds on.

Released

Sep 18, 2024

$ / min

—

Big Bench Audio

4.4%

Deepslate's proprietary end-to-end speech-to-speech model, built in Germany and hosted in the EU or on-premises for GDPR-sensitive voice agents.

Released

Undated

$ / min

$0.10/min

Big Bench Audio

85%

Alibaba's previous-generation full-duplex realtime voice model family (Plus and Flash) on Model Studio. It is still listed, but no longer recommended.

Released

Undated

$ / min

—

Big Bench Audio

99.2%

The realtime audio and video conversation variants of Alibaba's Qwen3.5-Omni, in Plus and Flash tiers, on Model Studio.

Released

Undated

$ / min

—

Big Bench Audio

98.7%

KRAFTON's open-weight full-duplex speech language model for real-time English conversation, with natural turn-taking, backchannels and barge-in. It's built on the Raon-Speech base model.

Released

Undated

$ / min

—

Big Bench Audio

57.5%

StepFun's full-duplex voice model on its Realtime API. It reasons while it speaks and runs tool calls in the background without breaking off the conversation.

Released

Undated

$ / min

$0.13/min

Big Bench Audio

99.7%