SpaceXAI just released Grok Voice 3.0. Grok Voice 3.0 is a SpaceXAI speech-to-speech model that listens, reasons, calls tools and speaks. It now ranks first on the Artificial Analysis Speech-to-Speech Index with a score of 86.9. For developers, the endpoint, events and $0.08 per minute price are unchanged from Grok Voice 2.0.
TL;DR
- Runs on: SpaceXAI’s hosted Voice Agent API only. No open weights, no local or quantized variants.
- Performance: Highest Index score of any model Artificial Analysis currently ranks.
- Best: 86.9 on the AA Speech-to-Speech Index, ranked #1.
- Worst: 71.6% on τ-Voice, behind GPT-Live-1 (Astra, medium) at 74.5%.
- Bottom line:
- Best: top composite score at 0.81 s time to first audio.
- Worst: not the leader on agentic task resolution or raw speech reasoning.
What is Grok Voice 3.0?
Grok Voice 3.0 is SpaceXAI’s flagship realtime voice model, exposed as grok-voice-3.0 on the /v1/realtime WebSocket endpoint. The grok-voice-latest alias already points to it, so existing agents switched automatically.
SpaceXAI has not published parameter counts or architecture details. Reasoning is on by default (reasoning.effort: "high") and can be set to "none" for faster turns. Server-side voice activity detection handles barge-in. Tools run inside the session: collections search, web search, X search, remote MCP servers and custom functions.
How does Grok Voice 3.0 perform on benchmarks?
The AA Speech-to-Speech Index is an equal-weighted average of 4 components. Those are Big Bench Audio reasoning, τ-Voice agentic performance, Speech Agent Arena preference and task success rate.
Grok Voice 3.0 scores 86.9. GPT-Live-1 (Astra, medium) scores 83.2, Gemini 3.8 Live Extended Thinking 82.6. Grok Voice 2.0 scored 81.3.
Component results are strong, not uniformly first:
- Big Bench Audio: 98.5%, ranked 4th. StepAudio 3 Realtime leads at 99.7%.
- Full Duplex Bench: 97.0%, ranked 4th. StepAudio 3 Realtime leads at 98.9%.
- Task success rate: 94.8%, up from 94.6% for Grok Voice 2.0.
- Speech Agent Arena Elo: 1070, the highest among the 4 models compared here.
- τ-Voice: 71.6%, up from 56.5%, but below GPT-Live-1 (Astra, medium) at 74.5%.
ServiceNow’s EVA-Bench scores enterprise voice agents on accuracy and experience. SpaceXAI reports a top accuracy score of 74%.
Is Grok Voice 3.0 fast and cheap?
Artificial Analysis measures 0.81 s time to first audio. GPT-Live-1 (Astra, medium) measures 1.34 s, so SpaceXAI’s “40% lower latency” claim checks out. Grok Voice 2.0 was faster, though, at 0.70 s.
On cost, SpaceXAI cites Artificial Analysis cost per Big Bench Audio task. Grok Voice 3.0 costs $0.025 per task, versus $0.030 for GPT-Live-1 (Astra, medium) and $0.055 for GPT-Realtime-2.1 (High). On AA’s cost per hour of input audio, Gemini 3.8 Live Extended Thinking is cheaper: $3.50 versus $4.80.
List prices are not directly comparable. GPT-Live-1 costs $0.05 per minute, but backend model and tool usage is billed separately. Gemini 3.8 Live bills audio per minute plus text tokens.
‘+s[0]+’
‘;
b.addEventListener(‘click’,function(){stop();show(i)});F.appendChild(b);});
function show(i){cur=(i+S.length)%S.length;F.querySelectorAll(‘.st’).forEach(function(x,j){x.classList.toggle(‘on’,j===cur)});
document.getElementById(‘gvDt’).textContent=S[cur][0];document.getElementById(‘gvDp’).textContent=S[cur][1];}
function stop(){if(timer){clearInterval(timer);timer=null;document.getElementById(‘gvPlay’).textContent=”Auto-play”;}}
document.getElementById(‘gvPrev’).onclick=function(){stop();show(cur-1)};
document.getElementById(‘gvNext’).onclick=function(){stop();show(cur+1)};
document.getElementById(‘gvPlay’).onclick=function(){if(timer){stop();return}this.textContent=”Pause”;timer=setInterval(function(){show(cur+1)},2200)};
show(0);
/* bars */
var M=[‘Grok Voice 3.0′,’Grok Voice 2.0′,’GPT-Live-1 (Astra, medium)’,’Gemini 3.8 Live Ext. Thinking’];
var B={
‘AA Index’:{v:[86.9,81.3,83.2,82.6],max:100,f:function(x){return x.toFixed(1)},n:’Equal-weighted average of Big Bench Audio, τ-Voice, Arena preference and task success.’},
‘Big Bench Audio’:{v:[98.5,97.2,90.1,97.7],max:100,f:function(x){return x.toFixed(1)+’%’},n:’Speech reasoning. Grok Voice 3.0 ranks 4th overall; StepAudio 3 Realtime leads at 99.7%.’},
‘τ-Voice’:{v:[71.6,56.5,74.5,68.6],max:100,f:function(x){return x.toFixed(1)+’%’},n:’Replica customer-support calls with tools. GPT-Live-1 (Astra, medium) leads this group.’},
‘Full Duplex Bench’:{v:[97.0,95.1,94.9,91.9],max:100,f:function(x){return x.toFixed(1)+’%’},n:’Pauses, turn-taking, interruptions and backchannels.’},
‘Task success’:{v:[94.8,94.6,87.4,89.1],max:100,f:function(x){return x.toFixed(1)+’%’},n:’Share of conversations ending in the correct final tool call.’},
‘Arena Elo’:{v:[1070,1011,1056,993],max:1100,f:function(x){return String(x)},n:’Live Speech Agent Arena Elo on Artificial Analysis. Higher is preferred.’},
‘Time to first audio’:{v:[0.81,0.70,1.34,1.35],max:1.5,f:function(x){return x.toFixed(2)+’ s’},n:’Lower is faster. Grok Voice 2.0 was quicker than 3.0.’}};
var C=document.getElementById(‘gvChips’),BR=document.getElementById(‘gvBars’),key=’AA Index’;
Object.keys(B).forEach(function(k){var c=document.createElement(‘button’);c.className=”chip”+(k===key?’ on’:”);c.textContent=k;
c.onclick=function(){key=k;C.querySelectorAll(‘.chip’).forEach(function(x){x.classList.toggle(‘on’,x===c)});drawBars()};C.appendChild(c);});
M.forEach(function(m,i){var r=document.createElement(‘div’);r.className=”row”;
r.innerHTML=’
‘+m+’
‘;BR.appendChild(r);});
function drawBars(){var d=B[key];BR.querySelectorAll(‘.row’).forEach(function(r,i){var f=r.querySelector(‘.fill’);f.style.width=”0″;
r.querySelector(‘.val’).textContent=d.f(d.v[i]);setTimeout(function(){f.style.width=(d.v[i]/d.max*100)+’%’},40+i*90)});
document.getElementById(‘gvNote’).textContent=d.n+’ Source: Artificial Analysis / SpaceXAI.’;}
drawBars();
/* scatter */
var P=[[‘Grok Voice 3.0′,’0.81 s · 71.6%’,0.81,71.6,1,’r’],[‘Grok Voice 2.0′,’0.70 s · 56.5%’,0.70,56.5,0,’r’],[‘GPT-Live-1 (Astra, medium)’,’1.34 s · 74.5%’,1.34,74.5,0,’l’],[‘Gemini 3.8 Live Ext. Thinking’,’1.35 s · 68.6%’,1.35,68.6,0,’l’]];
var G=document.getElementById(‘gvDots’),NS=’http://www.w3.org/2000/svg’;
P.forEach(function(p){var x=70+(p[2]-0.5)*540,y=280-(p[3]-50)/30*250;var g=document.createElementNS(NS,’g’);g.setAttribute(‘class’,’dot’);
var c=document.createElementNS(NS,’circle’);c.setAttribute(‘cx’,x);c.setAttribute(‘cy’,y);c.setAttribute(‘r’,p[4]?9:7);
c.setAttribute(‘fill’,p[4]?’#ffffff’:’#5c5c5c’);c.setAttribute(‘stroke’,p[4]?’#ffffff’:’#9a9a9a’);g.appendChild(c);
[[p[0],0,p[4]?’700′:’600′],[p[1],15,’500′]].forEach(function(L){var t=document.createElementNS(NS,’text’);
t.setAttribute(‘x’,p[5]===’r’?x+14:x-14);t.setAttribute(‘y’,y-2+L[1]);t.setAttribute(‘text-anchor’,p[5]===’r’?’start’:’end’);
t.setAttribute(‘fill’,p[4]?’#ffffff’:’#bdbdbd’);t.setAttribute(‘font-size’,’12’);t.setAttribute(‘font-weight’,L[2]);t.textContent=L[0];g.appendChild(t);});
G.appendChild(g);});
function drawDots(){G.querySelectorAll(‘.dot’).forEach(function(g,i){g.style.opacity=’0′;g.style.transform=’translateY(30px)’;
setTimeout(function(){g.style.opacity=’1′;g.style.transform=’none’},80+i*150)});}
window.addEventListener(‘load’,post);window.addEventListener(‘resize’,post);setTimeout(post,300);
})();
