grok-voice-3.0Grok Voice 3.0 is xAI's speech-to-speech model for voice agents, released on 14 October 2026. Speech goes in and spoken audio comes back from a single model, rather than through separate speech recognition, language and text-to-speech steps. xAI's announcement cites Artificial Analysis on 13 October 2026: first place on the Speech-to-Speech Index with a score of 86.9. Against Grok Voice 2.0, third-party coverage reports the index up from 81.3 to 86.9, the tau-Voice support-task benchmark from 56.5% to 71.6% and full-duplex turn-taking from 95.1% to 97.0%; time to first audio moved the other way, from 0.70 s to 0.81 s. On apimodels.app you open a WebSocket to wss://api.apimodels.app/v1/realtime?model=grok-voice-3.0 with your API key, or mint a short-lived token on your server so a browser can connect without ever seeing the key. The events are xAI's Voice Agent API, largely compatible with the OpenAI Realtime event model. Audio defaults to 16-bit PCM at 24 kHz; the 28 built-in voices each speak every supported language, and the model covers 20+ languages with automatic detection, Chinese included. Function calling carries no extra fee. The price is $0.104 per minute of connected session, billed per second, so a 90-second call costs $0.156, plus $0.0052 for each text message you send. One key covers this and 150+ other models, there is no separate xAI account or foreign card to set up, and every session appears in your call records with its duration and cost.
$0.104/min, billed per second · typed messages $0.0052 each · microphone required
Grok Voice 3.0
xAI cites Artificial Analysis (13 Oct 2026): first place with a score of 86.9, up from 81.3 for Grok Voice 2.0
Against Grok Voice 2.0, third-party coverage reports tau-Voice 56.5% to 71.6% and full-duplex turn-taking 95.1% to 97.0%; time to first audio went from 0.70 s to 0.81 s
wss://api.apimodels.app/v1/realtime with your API key on the server, or a short-lived token minted by your server for the browser
Every built-in voice speaks every supported language, Chinese included, with automatic language detection
Declare function tools in session.update. Server-side tools (web_search, x_search, file_search, mcp) are not available here yet
A 90-second session costs $0.156, each text message $0.0052, and unused seconds of the last minute are refunded
Grok Voice 3.0 is an Audio & Speech API provided by xAI. Grok Voice 3.0 is xAI's speech-to-speech model for voice agents, released on 14 October 2026. Speech goes in and spoken audio comes back from a single model, rather than through separate speech recognition, language and text-to-speech steps. xAI's announcement cites Artificial Analysis on 13 October 2026: first place on the Speech-to-Speech Index with a score of 86.9. Against Grok Voice 2.0, third-party coverage reports the index up from 81.3 to 86.9, the tau-Voice support-task benchmark from 56.5% to 71.6% and full-duplex turn-taking from 95.1% to 97.0%; time to first audio moved the other way, from 0.70 s to 0.81 s. On apimodels.app you open a WebSocket to wss://api.apimodels.app/v1/realtime?model=grok-voice-3.0 with your API key, or mint a short-lived token on your server so a browser can connect without ever seeing the key. The events are xAI's Voice Agent API, largely compatible with the OpenAI Realtime event model. Audio defaults to 16-bit PCM at 24 kHz; the 28 built-in voices each speak every supported language, and the model covers 20+ languages with automatic detection, Chinese included. Function calling carries no extra fee. The price is $0.104 per minute of connected session, billed per second, so a 90-second call costs $0.156, plus $0.0052 for each text message you send. One key covers this and 150+ other models, there is no separate xAI account or foreign card to set up, and every session appears in your call records with its duration and cost. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: per minute, billed per second: $$0.104 / min, per text message (input_text): $$0.0052.
Generate professional-grade voiceovers for videos, animations, and ads with diverse voice options.
Quickly produce podcast audio content with support for multi-character dialogue.
Convert text content into natural, fluid speech for audiobook production.
AI-powered multilingual dubbing and translation to help content reach global audiences.
Grok Voice 3.0 is available through APIMODELS at: per minute, billed per second: $$0.104 / min, per text message (input_text): $$0.0052. Billing is pay-as-you-go — you only pay for what you generate.
Sign up at APIMODELS, get your API key, and call our unified API endpoint. We provide detailed API documentation with code examples in cURL, Python, and Node.js.
APIMODELS offers the same Grok Voice 3.0 model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
Grok Voice 3.0 is xAI's speech-to-speech model for realtime voice agents, released on 14 October 2026: you stream speech in and the same model answers with speech. xAI's announcement cites Artificial Analysis (13 October 2026) placing it first on the Speech-to-Speech Index with a score of 86.9. Third-party coverage compares it with Grok Voice 2.0: the index rose from 81.3 to 86.9, the tau-Voice support-task benchmark from 56.5% to 71.6% and full-duplex turn-taking from 95.1% to 97.0%, while time to first audio went from 0.70 s to 0.81 s. On apimodels.app the model id is grok-voice-3.0.
$0.104 per minute of connected session time, billed per second, so a 90-second session costs $0.156. Each text message you send with conversation.item.create (input_text) adds $0.0052, and function calling has no extra fee. To connect you need at least one minute of balance ($0.104); each minute is charged when it starts and the unused seconds are refunded when the session ends. If your balance runs out mid-session you receive an error event of type insufficient_balance and the session closes; if the connection to the model fails, nothing is charged. The microphone demo on this page bills at the same rate.
From a server, open wss://api.apimodels.app/v1/realtime?model=grok-voice-3.0 with the header Authorization: Bearer YOUR_API_KEY. Browsers cannot set WebSocket headers, so your server first calls POST https://api.apimodels.app/v1/realtime/client_secrets with that same header (optional body {"expires_after":{"seconds":60}}, 10 to 600 seconds, default 60), receives {"value", "expires_at"}, and hands value to the page, which connects with new WebSocket(url, ["xai-client-secret." + value]). Never ship the API key to the browser. The endpoint exists only on api.apimodels.app, not under apimodels.app/api/v1, because WebSockets are not proxied there.
By default audio in both directions is 16-bit PCM, 24 kHz, mono, little-endian, base64-encoded inside JSON events; audio/pcmu, audio/pcma and audio/opus are also accepted through audio.input.format and audio.output.format, and audio.output.speed ranges from 0.7 to 1.5. There are 28 built-in voices, set with voice in session.update. Female: ara, aurora, carina, celeste, eve (the default), iris, liora, luna, ursa. Male: altair, atlas, castor, cosmo, helios, helix, kepler, leo, lumen, lux, naksh, orion, perseus, rex, rigel, sal, sirius, zagan, zenith.
Yes. It covers 20+ languages with native accents and detects the language automatically, and Chinese (Simplified) is on xAI's language list. Every built-in voice speaks every supported language, so there is no separate Chinese voice to choose. When you already know the caller's language, set audio.input.transcription.language_hint to a BCP-47 tag such as zh-CN.
Function calling, yes, at no extra fee: declare function tools in session.update, wait for response.function_call_arguments.done, run the function in your own code and send the result back as a function_call_output item with conversation.item.create, followed by response.create. Server-side tools (web_search, x_search, file_search and mcp) are not available on apimodels.app yet; a session.update that includes them gets an error event.
A session can last up to 60 minutes, an account can hold up to 5 concurrent sessions, and a session with no client events for 10 minutes is closed. Each session appears as one record in your console call history with its duration and cost, so you can reconcile usage call by call.
The event model is largely the same, so OpenAI Realtime client code ports with small changes: session.update, conversation.item.create, response.create and the input_audio_buffer events work the same way, and audio arrives as response.output_audio.delta. The differences: input transcription comes as cumulative conversation.item.input_audio_transcription.updated events rather than incremental deltas, and some OpenAI events are never emitted, namely conversation.item.retrieve, the output audio buffer started, stopped and cleared events, and rate_limits.updated. Billing differs too: one rate per minute of session here, where OpenAI bills audio input and output tokens separately.
On APIMODELS, Grok Voice 3.0 runs alongside 60+ models on one API key and one balance, so choosing is about fit, not lock-in. It supports Speech-to-Speech, Realtime WebSocket, 28 Voices, 20+ Languages, Function Calling, Billed per Second, and you can weigh it on price and capability against other Audio & Speech models, then switch by changing a single model-name string — no new account or integration. Browse every Audio & Speech option with live pricing at apimodels.app/models.
Grok Voice 3.0 supports: Speech-to-Speech, Realtime WebSocket, 28 Voices, 20+ Languages, Function Calling, Billed per Second. See the APIMODELS docs for full parameters and call examples.
Yes. APIMODELS exposes Grok Voice 3.0 through a single unified API and one key — no separate provider accounts, and no need to handle each provider's regional network access yourself.
We support Stripe (Visa, Mastercard, and other international cards) and Alipay. Credits are available instantly after payment.
How to get access, regional availability, and how this model compares with its alternatives.