Grok voices Aurora and Liora have quietly appeared on xAI’s official flagship voice roster, and the way they arrived is almost as interesting as the voices themselves. There was no press release, no dated changelog entry and no post on X announcing them. Instead, the two names simply showed up on the company’s “21 New Flagship Grok Voices” announcement page — a page originally published on 6 July 2026 with a different list of names.
We verified this directly on 24 August 2026. The live official page at x.ai now lists Aurora third and Liora fourth in its new-voice line-up, while third-party write-ups of the same announcement from early July list twenty-one names that include neither. That snapshot comparison is the whole story: somewhere between early July and late August, xAI expanded its Grok voices catalogue again and told nobody.
This article separates what is verified about the new Grok voices from what is not, sets the addition in the context of a summer in which the roster jumped from five voices to more than twenty-six, and compares xAI’s line-up with the voice options offered by OpenAI, Google, Anthropic and Microsoft. For businesses evaluating voice interfaces, the quiet cadence of these updates matters as much as the names.
Table of contents
- Aurora and Liora Join the Grok Voices Roster
- What We Know About Aurora and Liora — and What We Do Not
- From Five Grok Voices to 26 in One Summer
- Where the New Grok Voices Are Available Today
- The Voice Stack Behind the Roster: Think Fast 2.0 and Grok 4.6
- How Grok Voices Compare With Rival AI Assistants
- Why Voice Rosters Matter: The Voice AI Market in Numbers
- What the New Grok Voices Mean for Businesses
- References
Aurora and Liora Join the Grok Voices Roster
The evidence for the new Grok voices comes from comparing the official announcement page with dated snapshots of the same content.
What the official page shows now
xAI’s announcement page “21 New Flagship Grok Voices”, dated 6 July 2026, currently opens its voice list with Naksh and Atlas, followed immediately by Aurora and Liora, then Carina, Zagan, Helix, Orion and Luna. We fetched the page multiple times on 24 August 2026 and the two names were present on every fetch. Because the page presents them inside the flagship catalogue, Aurora and Liora carry the same headline properties xAI claims for the whole set: multilingual output and availability through the company’s realtime voice APIs.
How we know they were added later
TestingCatalog covered the announcement on 7 July 2026 and printed the full list of twenty-one names: Carina, Zagan, Helix, Orion, Luna, Iris, Altair, Zenith, Perseus, Helios, Lux, Kepler, Rigel, Cosmo, Celeste, Ursa, Sirius, Lumen, Castor, Naksh and Atlas. Aurora and Liora are not on it. Two independent third-party guides from July also omit both names. The live official page carrying them today therefore implies xAI slotted the pair into the roster at some point after 7 July 2026 — but no source states an exact date, so we will not invent one.
That pattern fits how xAI has behaved all year. New Grok voices, models and tools tend to land on console pages and documentation first, with formal announcements arriving late or never — the same quiet-shipping habit we documented when the mysterious Ox Alpha model surfaced on an evaluation platform without any attribution.
What We Know About Aurora and Liora — and What We Do Not
Honesty about the limits of the evidence matters more on a story like this than on a conventional launch, so here is the full accounting.
Verified facts
Aurora and Liora are listed on xAI’s official flagship voice page as of 24 August 2026. The flagship set they belong to is described by xAI as multilingual, cast for jobs such as support, characters, commentary, advertising and education, and available in the realtime Voice Agent API and the Text to Speech API. Those platform claims are stated on the page for the flagship catalogue as a whole.
What remains unconfirmed
No per-voice description exists for either name. Other entries on the page carry a category and a short character sketch — Carina, for instance, is filed under “Support” and described as “Soft, empathetic, and soothing” — but repeated extraction attempts could not attach any such text to the Aurora or Liora rows. Tone, gender presentation, accent and intended use case are all unknown, and any article claiming otherwise is guessing.
One further caution: Aurora is also the name xAI gave its image-generation model in December 2024. The Aurora in this story is a voice in the Grok voices catalogue, not that model, and the shared name is coincidence until xAI says otherwise.
From Five Grok Voices to 26 in One Summer
The quiet addition only makes sense against the July expansion that preceded it. On 6 July 2026, xAI published the announcement that transformed its catalogue. The company wrote: “Today, we’re releasing 21 new flagship voices for Grok Voice, joining the original five. All of them are multilingual and available now in the realtime Voice Agent API, the Text to Speech API, and the new Grok Voice Agent Builder.”
That single release took the roster of Grok voices from five to twenty-six — TestingCatalog framed it as “from five to 26” — and with Aurora and Liora now visible on the same page, the honest count is twenty-six plus. xAI also said the original five voices, Ara, Eve, Leo, Rex and Sal, were retrained so that “pacing, phrasing, and emphasis are more natural”, and that every flagship voice supports all of Grok Voice’s 25+ languages.
The same announcement introduced voice cloning: xAI says you can clone your own voice from about a minute of audio, and TestingCatalog notes custom voice clips can run up to 120 seconds. Taken together, the summer programme reads as a deliberate build-out of a voice platform, not a novelty feature — and the unannounced arrival of two more Grok voices in August suggests the build-out has not stopped.
Consumer app users, meanwhile, still see a much smaller menu. As of an early-August guide, the Grok app’s voice picker offers six voices with short personality labels:
| App voice | Listed character | In the original five? |
|---|---|---|
| Ara | Upbeat female | Yes |
| Eve | Soothing female | Yes |
| Leo | British male | Yes |
| Rex | Calm male | Yes |
| Sal | Smooth male | Yes |
| Gork | Lazy male | No — app-only addition |
How the voice programme got here
The roster expansion caps eighteen months of steady voice investment. Grok’s voice mode first arrived in late February 2025, initially on iOS behind the Premium+ paywall, and early coverage highlighted its unusual personality settings — modes reported at launch included “unhinged”, “romantic” and “genius”, a deliberately looser register than rivals allowed.
In July 2025 xAI added companions, animated 3D characters users can talk with; by late February 2026, Wikipedia records five of them: Good Rudi, Bad Rudi, Ani, Mika and Valentine. Personality modes, companions and now a castable catalogue of Grok voices are three distinct layers of the same strategy: making the assistant something you talk to rather than type at.
Where the New Grok Voices Are Available Today
The official platform list is specific, and it is worth reading precisely because it does not mention the consumer app.
The API surfaces
According to xAI, the flagship Grok voices — the set Aurora and Liora now sit in — are available in the realtime Voice Agent API, the Text to Speech API and the Grok Voice Agent Builder, the console tool we examined when xAI introduced its Voice Agent Builder. TestingCatalog adds concrete limits and prices: text-to-speech accepts up to 15,000 characters per request, audio generation costs $0.05 per minute of audio, and telephony integration runs at $0.01 per minute.
The consumer app question
Whether app users will ever get the new Grok voices is genuinely unresolved. The August guide that lists the app’s six voices states plainly that none of the twenty-one July voices shipped to the consumer app’s voice picker. At least one aggregator has claimed the announcement confirms availability “across iOS, Android, and web”, but the official page says no such thing — it names the APIs and the builder tool. Until xAI’s app changelog says otherwise, treat the new voices, Aurora and Liora included, as developer-facing.
What Grok voice services cost
| Service | Verified price | Source |
|---|---|---|
| Text to Speech audio | $0.05 per minute of audio | xAI / TestingCatalog |
| Telephony integration | $0.01 per minute | TestingCatalog |
| Grok Voice Think Fast 2.0 | $0.08 per minute | TestingCatalog |
| Grok 4.6 tokens | $2 per million input, $6 per million output | TestingCatalog |
The Voice Stack Behind the Roster: Think Fast 2.0 and Grok 4.6
A bigger catalogue of Grok voices would matter little if the underlying speech stack were weak, and August 2026 brought upgrades on that front too.
The speech-to-speech engine
xAI’s voice system is an in-house speech-to-speech stack rather than a licensed third-party engine, sitting alongside the company’s wider work in natural language processing. In its December 2025 Voice Agent API announcement, xAI claimed an average time-to-first-audio of under one second — “nearly 5 times faster than the closest competitor” — a number-one ranking on the Big Bench Audio benchmark at 95%, and deployment at scale: the same stack powers Grok Voice “for millions in our mobile apps and Tesla vehicles”.
The current generation is Grok Voice Think Fast 2.0, announced on 29 July 2026 as the successor to Grok Voice Think Fast 1.0, which we reviewed earlier this year. On 5 August 2026 the grok-voice-latest alias switched automatically from version 1.0 to 2.0, so every integration riding the alias — the new Grok voices included — now runs on the upgraded model.
The version 2.0 gains are stated precisely: the overall voice benchmark score rose from 75.7% to 82.9%, agentic performance rose from 52.1% to 56.5%, and time to first audio fell from 1.25 seconds to 0.70 seconds — a drop of 0.55 seconds, or roughly 44%. xAI evaluated the model across 24 languages and claims transcription accuracy 1.5 to 2.0 times better than named rivals.
The model behind the answers
The words the Grok voices speak come from the text model, and that was refreshed in August as well. Grok 4.6, released on 12 August 2026, scored 61 on the Artificial Analysis Intelligence Index, up from 56 for Grok 4.5, at $2 per million input tokens and $6 per million output tokens. A faster brain plus a faster mouth is the combination xAI is clearly assembling.
How Grok Voices Compare With Rival AI Assistants
Voice roster size is a crude metric, but it is one of the few that can be compared across vendors with verified numbers, and on it the Grok voices catalogue now leads by a wide margin.
The verified numbers behind the chart
OpenAI’s Advanced Voice Mode reached nine voices in September 2024, when Arbor, Maple, Sol, Spruce and Vale joined Breeze, Juniper, Cove and Ember. Google’s Gemini Live launched with ten voices in August 2024. Anthropic’s Claude voice mode arrived in beta in May 2025 with five voice options, English-only at launch. Microsoft has promoted four new Copilot voices. Each figure comes from the vendor’s own material or contemporaneous reporting, and each is a snapshot — rosters move.
| Assistant | Voices | Verified context |
|---|---|---|
| Grok Voice | 26+ in the API roster | 21 added July 2026; Aurora and Liora appeared by late August; app picker still shows 6 |
| ChatGPT Advanced Voice | 9 | Five voices added September 2024 |
| Gemini Live | 10 at launch | Named voices, August 2024 |
| Claude voice mode | 5 | Beta from May 2025, mobile apps |
| Microsoft Copilot | 4 | Four new voices promoted in 2025 |
A fair caveat: raw counts flatten real differences. A catalogue built for developer APIs, like the Grok voices roster, naturally grows faster than a consumer picker that each user scrolls through. The sharper comparison is strategic — xAI is selling voices as infrastructure, while most rivals still treat them as a personalisation setting.
Why Voice Rosters Matter: The Voice AI Market in Numbers
The commercial backdrop explains why every major lab is suddenly investing in voice casts. Industry statistics compiled in 2026 put hard numbers on the shift. Market.us projects the voice AI agents market to grow from $2.4 billion in 2024 to $47.5 billion by 2034, a 34.8% compound annual growth rate — close to a twenty-fold expansion. Grand View Research separately sizes conversational AI at $11.58 billion in 2024, heading for $41.39 billion by 2030 at a 23.7% CAGR.
Two further figures sharpen the picture. Gartner expects conversational AI to cut contact-centre agent labour costs by $80 billion in 2026, and Nextiva reports that 80% of businesses plan to integrate AI-driven voice technology into customer service by 2026. If those forecasts are even directionally right, a deep bench of distinct, castable voices becomes a genuine competitive asset — which is precisely how xAI’s announcement frames its roster, with voices “cast for a specific job: support, characters, commentary, advertising, education”.
Latency numbers feed the same argument. Sub-second time to first audio is the difference between a voice agent that feels like a phone menu and one that feels like a colleague, and xAI’s stated 0.70 seconds is squarely aimed at the customer-facing use cases those market forecasts describe.
What the New Grok Voices Mean for Businesses
For most organisations the practical question is not which lab has the longest list, but what a fast-moving voice platform makes possible — and what it obliges you to check.
Opportunities worth testing
Voice agents for support lines, out-of-hours booking and internal helpdesks are now cheap enough to pilot: at $0.05 per minute of generated audio and $0.01 per minute of telephony, a thousand-minute month of caller conversation costs tens of dollars, not thousands. The expanded Grok voices roster means a brand can cast different voices for different functions — one for support, another for marketing content — without commissioning studio recordings. Voice cloning from a minute of audio extends that to a company’s own recorded voice, with all the consent and governance questions that implies.
If your team is weighing up where voice fits in a wider automation programme, our AI strategy practice covers exactly this evaluation, from use-case selection through to vendor comparison.
A sensible pilot sequence is short. Pick one contained workflow — say, appointment confirmations — and script it end to end. Audition a shortlist of Grok voices against the same script, including at least one of the newer names, and have real staff rather than the project team judge the output. Measure completion rate and caller drop-off against your current process for a month before widening scope. The per-minute pricing means the experiment’s cost ceiling is known in advance, which is rarely true of AI projects.
Cautions before you commit
The same quiet-shipping habit that makes xAI interesting makes it hard to govern. Aurora and Liora arrived with no descriptions, no changelog and no dates; a voice you cast today could be retrained tomorrow, as the original five were in July. Pin specific voice identifiers in production rather than riding a latest alias, test after upgrades like the 5 August alias switch, and keep transcripts of what your agents say. And remember the platform split: the full roster of Grok voices lives in the APIs and the Agent Builder, not the consumer app your staff may be trialling on their phones.
The two new names are a small story on their own. As evidence of how fast conversational AI platforms now evolve — twenty-plus voices, a model upgrade and a silent roster edit inside eight weeks — the Grok voices catalogue is one of the clearest signals yet that voice has moved from demo to product.
References
21 New Flagship Grok Voices — xAI
xAI Launches 21 New Grok Voices on xAI Console — TestingCatalog
xAI Launches Grok Voice Think Fast 2.0 on Agent Builder — TestingCatalog
xAI Releases Grok 4.6 for Long-Running Agent Work — TestingCatalog
Grok Voice Mode: Complete Guide — Fello AI
OpenAI Rolls Out Advanced Voice Mode With More Voices — TechCrunch
Anthropic Launches a Voice Mode for Claude — TechCrunch
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.