Primary use stated or inferred cautiously from official documentation.
X
XAI / grok-voice-3.0
Grok Voice 3.0
Proprietary voice model available since October 14, 2026 for bidirectional audio and text conversations. Its versioned identifier is grok-voice-3.0; grok-voice-latest is an alias already pointing to this version, not another model. It replaces the deprecated Grok Voice 2.0 without changing the endpoint, session parameters or events. Do not confuse it with Grok Voice Transcribe 2.0 or the separate text-to-speech service.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | grok-voice-3.0 |
|---|---|
| Model type | Voice and audio |
| Access model | Paid proprietary |
| License | Proprietary |
| Deployment | Hosted API |
| Release | 14 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Audio |
| Exit | Text · Audio |
| Context window | Context window not published |
| maximum output | Streaming audio and text; maximum not published |
| Reasoning | Effort high or none; high by default |
| Published tools | Features · Web search · Search on X · MCP · Collections search |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | xAI API |
Spoken assistants, telephone agents and multilingual interaction with tools. Documentation lists Spanish, English, French, Italian, German and Portuguese among more than twenty languages. It supports automatic turn detection, functions and configured access to search or MCP. Tool support does not authorize sensitive actions: retain human confirmation where appropriate.
Limits concern concurrent sessions, not tokens: tiers T0, T1, T2, T3 and T4 allow 10, 20, 50, 100 and 200 sessions respectively; check your team limits. We have not verified a public context window or maximum duration. Advertised latency is not an independent benchmark. Use ephemeral tokens in browsers and never expose a permanent key. Avoid the latest alias when pinning a reproducible version.
03 / CAPABILITIES
What it can do
Orientation
Spoken assistants, telephone agents and multilingual interaction with tools. Documentation lists Spanish, English, French, Italian, German and Portuguese among more than twenty languages. It supports automatic turn detection, functions and configured access to search or MCP. Tool support does not authorize sensitive actions: retain human confirmation where appropriate.
Context
Context window not published · Streaming audio and text; maximum not published
Tools and integration
Functions · Web search · X search · MCP · Collections search
Access
xAI API · WebSocket /v1/realtime
Input modalities
TextAudio
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | USD 0.08 / conversation minute | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | USD 0.004 / text input | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
The official rate is USD 0.08 per conversation minute, equivalent to USD 4.80 per hour, and USD 0.004 per text input. Voice 2.0 pricing is unchanged. These are not rates per million tokens: numerical token-comparison fields remain empty. Separate transcription and speech-synthesis services use different units and are not imported into this profile.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
We have not located published numerical scores specific to Grok Voice 3.0. Inferama has not run its own evaluation or transferred results from Grok 4.7 or the transcription model. Useful testing should control language, accent, noise, interruptions, transcript fidelity and tool actions, while recording the exact identifier, observed latency and actual cost.
i We have not found published benchmarks for this model.
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Grok Voice 3.0
Available; replaces Voice 2.0 and updates the grok-voice-latest alias.
07 / SOURCES