# Grok Voice 3.0

> Source: Benchize (https://benchize.com) — https://benchize.com/models/grok-voice-3-0. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/models/grok-voice-3-0.

xAI

Canonical URL: https://benchize.com/models/grok-voice-3-0
Last updated: Thu, 22 Oct 2026 06:30:09 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

xAI

## Benchmarks

| Benchmark | Score | Rank |  |
| --- | --- | --- | --- |
| [τ-bench retail](https://benchize.com/benchmarks/sierra%3Aretail.md) | 76.32% | 5 / 8 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `pass_1` \| 76.31578947368422 \| percent \| consistent_success_across_k_trials \| Higher is better \| ### Configuration benchmarkVersion 4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699 harness tau-bench methodology {"evaluation_date":"2026-10-08","notes":"Sierra evaluation of an early-access Grok Voice 3.0 model before its public release. The public API model name is grok-voice-3.0. Standard full-duplex xAI adapter, base airline/retail/telecom task sets, regular speech complexity, seed 300, 0.2-second ticks, 1200-second simulated limit, and one scored trial per task. The voice user simulator v1.0 pipeline used gpt-5.5 with xhigh reasoning effort as its customer LLM; the remaining simulator components, benchmark prompts, policies, tools, and evaluator were unchanged. This non-default customer LLM makes the result a custom submission. Published trajectory metadata replaces an internal early-access model identifier with the public API model name; scored task content, rewards, timestamps, and audio are unchanged.","tau2_bench_version":"4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699","user_simulator":"gpt-5.5 (xhigh)","verification":{"details":"All 278 base tasks have one scored canonical trajectory. The initial retail run encountered ungraded connection timeouts at 80 concurrency and was resumed at 40 concurrency; all 114 retail tasks have a final scored result. Airline ran at up to 80 concurrency and telecom at 40. The submitted scores were computed from the saved trajectories. Reviewed trajectories and canonical audio are hosted by the leaderboard.","modified_prompts":false,"omitted_questions":false}} modality voice Reasoning effort unknown retrieval Not reported sourceSubmission grok-voice-3-0_xai_2026-10-14 ### Source [Original source](https://sierra-tau-bench-public.s3.us-west-2.amazonaws.com/submissions/grok-voice-3-0_xai_2026-10-14/submission.json) Source sierra Original model label Grok Voice 3.0 Evaluated 2026-10-14 Observed 2026-10-21T16:54:59.999Z Evidence SHA-256 `ba2fe869ab163c4df258ca8252f679484c0002ab1dc01e9bcc4e3059ea54f6e5` Source locator `/results/retail` |  |  |  |
| [τ-bench airline](https://benchize.com/benchmarks/sierra%3Aairline.md) | 80% | 6 / 8 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `pass_1` \| 80 \| percent \| consistent_success_across_k_trials \| Higher is better \| ### Configuration benchmarkVersion 4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699 harness tau-bench methodology {"evaluation_date":"2026-10-08","notes":"Sierra evaluation of an early-access Grok Voice 3.0 model before its public release. The public API model name is grok-voice-3.0. Standard full-duplex xAI adapter, base airline/retail/telecom task sets, regular speech complexity, seed 300, 0.2-second ticks, 1200-second simulated limit, and one scored trial per task. The voice user simulator v1.0 pipeline used gpt-5.5 with xhigh reasoning effort as its customer LLM; the remaining simulator components, benchmark prompts, policies, tools, and evaluator were unchanged. This non-default customer LLM makes the result a custom submission. Published trajectory metadata replaces an internal early-access model identifier with the public API model name; scored task content, rewards, timestamps, and audio are unchanged.","tau2_bench_version":"4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699","user_simulator":"gpt-5.5 (xhigh)","verification":{"details":"All 278 base tasks have one scored canonical trajectory. The initial retail run encountered ungraded connection timeouts at 80 concurrency and was resumed at 40 concurrency; all 114 retail tasks have a final scored result. Airline ran at up to 80 concurrency and telecom at 40. The submitted scores were computed from the saved trajectories. Reviewed trajectories and canonical audio are hosted by the leaderboard.","modified_prompts":false,"omitted_questions":false}} modality voice Reasoning effort unknown retrieval Not reported sourceSubmission grok-voice-3-0_xai_2026-10-14 ### Source [Original source](https://sierra-tau-bench-public.s3.us-west-2.amazonaws.com/submissions/grok-voice-3-0_xai_2026-10-14/submission.json) Source sierra Original model label Grok Voice 3.0 Evaluated 2026-10-14 Observed 2026-10-21T16:54:59.999Z Evidence SHA-256 `ba2fe869ab163c4df258ca8252f679484c0002ab1dc01e9bcc4e3059ea54f6e5` Source locator `/results/airline` |  |  |  |
| [τ-bench telecom](https://benchize.com/benchmarks/sierra%3Atelecom.md) | 82.46% | 7 / 8 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `pass_1` \| 82.45614035087719 \| percent \| consistent_success_across_k_trials \| Higher is better \| ### Configuration benchmarkVersion 4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699 harness tau-bench methodology {"evaluation_date":"2026-10-08","notes":"Sierra evaluation of an early-access Grok Voice 3.0 model before its public release. The public API model name is grok-voice-3.0. Standard full-duplex xAI adapter, base airline/retail/telecom task sets, regular speech complexity, seed 300, 0.2-second ticks, 1200-second simulated limit, and one scored trial per task. The voice user simulator v1.0 pipeline used gpt-5.5 with xhigh reasoning effort as its customer LLM; the remaining simulator components, benchmark prompts, policies, tools, and evaluator were unchanged. This non-default customer LLM makes the result a custom submission. Published trajectory metadata replaces an internal early-access model identifier with the public API model name; scored task content, rewards, timestamps, and audio are unchanged.","tau2_bench_version":"4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699","user_simulator":"gpt-5.5 (xhigh)","verification":{"details":"All 278 base tasks have one scored canonical trajectory. The initial retail run encountered ungraded connection timeouts at 80 concurrency and was resumed at 40 concurrency; all 114 retail tasks have a final scored result. Airline ran at up to 80 concurrency and telecom at 40. The submitted scores were computed from the saved trajectories. Reviewed trajectories and canonical audio are hosted by the leaderboard.","modified_prompts":false,"omitted_questions":false}} modality voice Reasoning effort unknown retrieval Not reported sourceSubmission grok-voice-3-0_xai_2026-10-14 ### Source [Original source](https://sierra-tau-bench-public.s3.us-west-2.amazonaws.com/submissions/grok-voice-3-0_xai_2026-10-14/submission.json) Source sierra Original model label Grok Voice 3.0 Evaluated 2026-10-14 Observed 2026-10-21T16:54:59.999Z Evidence SHA-256 `ba2fe869ab163c4df258ca8252f679484c0002ab1dc01e9bcc4e3059ea54f6e5` Source locator `/results/telecom` |  |  |  |

---
Cite as: Grok Voice 3.0 — Benchize, https://benchize.com/models/grok-voice-3-0 (retrieved 2026-10-22)