Cartesia Ships Sonic-3.6: A Streaming TTS Model That Now Leads Both Artificial Analysis Speech Arenas

Cartesia has released Sonic-3.6, the newest version of its real-time text-to-speech model. It arrives roughly three months after Sonic-3.5. The new change is naturalness, and this one is independently checkable. Sonic 3.6 now holds #1 on both Artificial Analysis speech leaderboards — 1,283 Elo on the Provider Voice board and 1,123 on the Controlled Voice board. The second result matters more. That board clones every model onto the same eight reference voices, which isolates the synthesis engine from the voice catalog. Sonic-3.6 leads it, with Sonic-3.5 second and ElevenLabs Eleven v3 third. The model runs on state space models rather than transformers, and Cartesia states sub-90ms time-to-first-audio. It is available in beta.

Is it deployable?

YES, it is available in beta and as a hosted API. Not as self-hosted weights.

Sonic is a closed, commercial model. There are no open weights and no Hugging Face repo. You rent it.

Company level: Solo developers and startups (Free/Pro $5 tiers), scaleups running contact centers (Startup $49 / Scale $299), and regulated enterprises needing DPAs, BAAs, and SSO.
Industries: Financial services, healthcare, retail and e-commerce, logistics, recruiting, SaaS support, consumer companion apps, media localization
Applications: Inbound support agents, outbound qualification calls, IVR replacement, appointment reminders, sales-training simulators, audio localization, in-product voice UI

The Architecture

Sonic runs on state space models rather than transformers. Cartesia’s launch page frames the usual tradeoffs — speed versus naturalness, accuracy versus cost — as architectural, not inevitable.

The practical output is time-to-first-audio. Cartesia states sub-90ms TTS latency, and 100ms transcript latency for its Ink-2 speech-to-text model. Both are vendor-stated model latency, not measured end-to-end round trips.

Interactive explainer

<![CDATA[<![CDATA[<![CDATA['; host.appendChild(m); } function reset(){ if(raf){cancelAnimationFrame(raf);raf=null} t0=null; [].slice.call(document.querySelectorAll('.cell')).forEach(function(c){c.classList.remove('on')}); pbB.style.left="0"; pbA.style.left="0"; msB.innerHTML='first audio: —'; msA.innerHTML='first audio: —'; clock.textContent="0 ms"; } function step(ts){ if(!t0) t0=ts; var el=(ts-t0)*0.55; // slow it down so it is readable if(el>TOTAL){ el=TOTAL; } clock.textContent=Math.round(el)+’ ms’; var pct=(el/TOTAL*100)+’%’; pbB.style.left=pct; pbA.style.left=pct; [].slice.call(document.querySelectorAll(‘.cell’)).forEach(function(c){ if(+c.dataset.at<=el) c.classList.add('on'); }); if(el>=B_FIRST) msB.innerHTML=’first audio: <b style="color:#48c96a">88 ms</b>‘; if(el>=A_FIRST) msA.innerHTML=’first audio: <b style="color:#d8a13a">430 ms</b>‘; if(el<TOTAL) raf=requestAnimationFrame(step); else raf=null; } document.getElementById('play').addEventListener('click',function(){reset();raf=requestAnimationFrame(step)}); document.getElementById('reset').addEventListener('click',reset); buildLanes(); reset(); /* ================= 2. arena ================= */ var BOARDS={ prov:{ base:1150, span:150, rows:[ ['Cartesia Sonic 3.6',1283,17,1], ['Alibaba Qwen-Audio-3.0-TTS-Plus',1240,14,0], ['SpeechifyAI Simba 3.2',1240,14,0], ['VUI Labs Luna TTS',1219,14,0], ['Google Gemini 3.1 Flash TTS',1211,12,0], ['Cartesia Sonic 3.5',1204,13,0], ['ElevenLabs Eleven v3',1179,11,0] ]}, ctrl:{ base:1000, span:140, rows:[ ['Cartesia Sonic 3.6',1123,16,1], ['Cartesia Sonic 3.5',1102,13,0], ['ElevenLabs Eleven v3',1059,13,0], ['StepFun StepAudio 2.5 TTS',1049,14,0], ['Inworld Realtime TTS 1.5 Max',1047,14,0], ['Inworld Realtime TTS-2 (RP)',1044,13,0], ['SpaceXAI TTS',1040,13,0] ]} }; var board='prov'; var barsEl=document.getElementById('bars'); function drawBars(){ var B=BOARDS[board]; barsEl.innerHTML=''; B.rows.forEach(function(r){ var pct=Math.max(6,Math.min(100,(r[1]-B.base)/B.span*100)); var ciW=(r[2]*2)/B.span*100; var d=document.createElement('div'); d.className="bar"+(r[3]?' win':''); d.innerHTML=' <div class="bmeta"><span class="nm">‘+r[0]+’</span><span class="el">‘+r[1].toLocaleString()+’ Elo</span></div> ‘+ ‘ <div class="btrack"> <div class="bfill"></div> ‘+ (r[3]?’ <div class="ci" style="left:'+Math.max(0,pct-ciW/2)+'%;width:'+ciW+'%"></div> ‘:”)+’</div> ‘; barsEl.appendChild(d); requestAnimationFrame(function(){ d.querySelector(‘.bfill’).style.width=pct+’%’; }); }); } [].slice.call(document.querySelectorAll(‘.toggle button’)).forEach(function(b){ b.addEventListener(‘click’,function(){ [].slice.call(document.querySelectorAll(‘.toggle button’)).forEach(function(x){x.classList.remove(‘on’)}); b.classList.add(‘on’); board=b.dataset.b; drawBars(); }); }); drawBars(); /* ================= 3. transcript ================= */ var S={laugh:0,pause:0,ipa:0,alnum:0,fast:0,loud:0,hinglish:0}; var tEl=document.getElementById(‘tscript’), pEl=document.getElementById(‘params’); var cv=document.getElementById(‘wave’), cx=cv.getContext(‘2d’); function render(){ var s=””; s+= S.hinglish ? ‘Aapka order confirm ho gaya hai, ‘ : ‘Your order is confirmed, ‘; if(S.ipa) s+=’and the <span class="ipa"><<s|ə|ˈ|p|i|n|ə>></span> notice was filed. ‘; if(S.alnum) s+=’Reference <b>A7-4419-KX</b>, callback <b>+1 415 555 0142</b>. ‘; if(S.pause) s+=’<span class="pause"></span>‘; s+= S.hinglish ? ‘so we are all set. ‘ : ‘so you are all set. ‘; if(S.laugh) s+=’<span class="tag">[laughter]</span> Finally!’; tEl.innerHTML=s; var p=[]; p.push(‘speed <b>‘+(S.fast?’1.4′:’1.0′)+’</b>‘); p.push(‘volume <b>‘+(S.loud?’1.6′:’1.0′)+’</b>‘); p.push(‘language <b>‘+(S.hinglish?’hi + en (code-switch)’:’en’)+’</b>‘); p.push(‘non-verbals <b>‘+(S.laugh?'[laughter]’:’none’)+’</b>‘); p.push(‘pronunciation dict <b>‘+(S.ipa?’IPA inline’:’off’)+’</b>‘); var out=””; p.forEach(function(x){ out+=’<span>‘+x+’</span>‘; }); pEl.innerHTML=out; drawWave(); } var phase=0; function drawWave(){ var w=cv.width,h=cv.height; cx.clearRect(0,0,w,h); var amp=(S.loud?0.80:0.52)*h/2; var freq=(S.fast?0.055:0.038); // baseline grid cx.strokeStyle=”rgba(143,163,146,.14)”; cx.lineWidth=1; cx.beginPath(); cx.moveTo(0,h/2); cx.lineTo(w,h/2); cx.stroke(); // bars var bars=Math.floor(w/6); for(var i=0;i<bars;i++){ var x=i*6+2, t=i*freq+phase; var env=Math.sin(i/bars*Math.PI); var v=(Math.sin(t)*0.5+Math.sin(t*2.3)*0.3+Math.sin(t*4.7)*0.2); var burst=1; if(S.laugh && i>bars*0.80) burst=1.7+Math.sin(t*9)*0.5; // laughter flourish if(S.pause && i>bars*0.46 && i<bars*0.56) burst=0.05; // silence gap var mag=Math.abs(v)*amp*env*burst+2; var g=cx.createLinearGradient(0,h/2-mag,0,h/2+mag); g.addColorStop(0,'#48c96a'); g.addColorStop(0.5,'#309D4B'); g.addColorStop(1,'#1d6b33'); cx.fillStyle=g; cx.fillRect(x,h/2-mag,3.4,mag*2); } } function loop(){ phase+=0.05; if(document.getElementById('p2').classList.contains('on')) drawWave(); requestAnimationFrame(loop); } [].slice.call(document.querySelectorAll('.chip')).forEach(function(c){ c.addEventListener('click',function(){ S[c.dataset.k]=S[c.dataset.k]?0:1; c.classList.toggle('on'); render(); }); }); render(); loop(); /* ================= 4. plan fit ================= */ var PLANS=[ {n:'Free', p:'$0', min:27, cc:2, note:'20K credits · no commercial licence'}, {n:'Pro', p:'$5', min:133, cc:3, note:'100K credits · commercial licence + instant cloning'}, {n:'Startup',p:'$49', min:1667, cc:5, note:'1.25M credits · pro voice cloning + organizations'}, {n:'Scale', p:'$299', min:10667, cc:15, note:'8M credits · priority support + high concurrency'}, {n:'Enterprise',p:'Custom',min:Infinity,cc:'custom',note:'DPAs/BAAs · SSO · custom concurrency'} ]; var sl=document.getElementById('sl'), slv=document.getElementById('slv'), slc=document.getElementById('slc'); var plansEl=document.getElementById('plans'), fitEl=document.getElementById('fit'); function mins(v){ // exponential feel: 5 -> 30000 return Math.round(5*Math.pow(30000/5,v/60)); } function paintPlans(){ var m=mins(+sl.value); slv.textContent=m.toLocaleString()+’ min’; slc.textContent=”\u2248 “+(m/60).toFixed(m<600?1:0)+' hours of generated speech per month'; var idx=PLANS.findIndex(function(p){return m<=p.min}); if(idx<0) idx=PLANS.length-1; plansEl.innerHTML=''; PLANS.forEach(function(p,i){ var d=document.createElement('div'); d.className="plan"+(i===idx?' fit':''); d.innerHTML=' <div class="pn">‘+p.n+’</div> <div class="pp">‘+p.p+’</div> ‘+ ‘ <div class="pm">‘+(p.min===Infinity?’custom’:’~’+p.min.toLocaleString()+’ min’)+’<br />‘+p.cc+’ concurrent</div> ‘; plansEl.appendChild(d); }); var P=PLANS[idx]; var extra = idx===0 ? ‘ You would still need <b>Pro</b> or higher before shipping anything commercial.’ : (idx===4 ? ‘ At this volume you are negotiating custom credits, concurrency and a DPA/BAA.’ : ”); fitEl.innerHTML=’At <b>‘+m.toLocaleString()+’ minutes/month</b> the fit is <b>‘+P.n+’</b> (‘+P.p+ (P.p===’Custom’?”:’/mo’)+’) — ‘+P.note+’.’+extra; } sl.addEventListener(‘input’,paintPlans); paintPlans(); /* ———- iframe auto-resize ———- */ function resize(){ try{ var h=document.body.offsetHeight+40; if(window.parent && window.parent!==window) window.parent.postMessage({mtpHeight:h},’*’); }catch(e){} } window.addEventListener(‘load’,resize); setTimeout(resize,300); setTimeout(resize,900); })(); </script> </body> </html> “>]]]]]]><![CDATA[><![CDATA[>]]]]><![CDATA[>]]>

Features that matter in production

Sonic exposes controls built for agent transcripts rather than narration:

Inline expression tags. Non-verbal expressions like [laughter] go directly in the transcript.
Instant voice cloning from about 10 seconds of audio.
Custom pronunciation dictionaries, including IPA overrides such as <> for subpoena.
Speed, volume, and emotion parameters exposed through the API and integrations like the LiveKit Agents plugin.
Native alphanumerics. Order numbers, phone numbers, and confirmation codes read correctly without preprocessing.

Cartesia’s launch demos show English with natural pauses and filler words, plus Hinglish code-switching between Hindi and English.

Apple’s macOS Tahoe Preview May Have Leaked A Ton Of Unreleased Products

LG Hosts NVIDIA at Seoul Robot Data Factory as 100,000-Hour Training Push Takes Shape – Unite.AI

Pricing reality

Artificial Analysis normalizes Sonic 3.6 at $49.00 per 1M characters. That is half of ElevenLabs Eleven v3 at $100.00, and well above Speechify Simba 3.2 at $10.00 for a 1,240 Elo.

Cartesia sells credits, not characters. Scale at $299 per month includes roughly 10,667 TTS minutes and 15 concurrent requests. Line voice agents bill separately at $0.06 per minute.

Key Takeaways

Sonic-3.6 is #1 on both Artificial Analysis speech arenas — 1,283 Elo Provider Voice, 1,123 Controlled Voice.
Winning the Controlled board means the engine improved, not just the voice catalog.
It is beta on Cartesia’s API only; docs still list Sonic 3.5 as stable, and partners carry 3.5.
Deployable as a hosted API, not self-hosted weights; commercial use starts at the $5 Pro tier.
Latency claims (sub-90ms TTFA) are vendor-stated model latency, so benchmark your own round trip.

Check out the Project Page-Cartesia Sonic, Cartesia launch page, Cartesia pricing, Cartesia docs, Artificial Analysis Speech Arena and @cartesia on X. All figures verified August 18, 2026.. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.

Credit: Source link

Cartesia Ships Sonic-3.6: A Streaming TTS Model That Now Leads Both Artificial Analysis Speech Arenas

YOU MAY ALSO LIKE

Apple’s macOS Tahoe Preview May Have Leaked A Ton Of Unreleased Products

LG Hosts NVIDIA at Seoul Robot Data Factory as 100,000-Hour Training Push Takes Shape – Unite.AI

Related Posts

Apple’s macOS Tahoe Preview May Have Leaked A Ton Of Unreleased Products

LG Hosts NVIDIA at Seoul Robot Data Factory as 100,000-Hour Training Push Takes Shape – Unite.AI

Nous Research Ships Bot Mode for Hermes Agent, Turning Agent Profiles Into a Roster of Named Bots

ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation

South Korea president calls for military independence in wake of Trump call to cut joint drills - The Guardian

Leave a Reply Cancel reply

Search

Steve Kornacki: Incumbents in New York, Maryland, Utah look to fend off primary challenges

Karsan’s Autonomous E-ATAK Starts Passenger Service at Efteling Theme Park – Unite.AI

Suspect goes on hit-and-run rampage after carjacking

About

Legal

Bloggers

Contact