Cartesia Sonic 3.6
Sonic 3.6 (27 août 2026) — 44 langues, plafond éditeur sous 90 ms. Arena publique 17 oct. : ELO 1285, rang 4.
Comparative Scores
Architecture
Still the low-latency cascade candidate for a French voice test, now on Sonic 3.6 rather than 3.5. Pair it with a French STT other than Ink-2. Treat on-prem as a contract to read, not as Swiss residency already obtained.
Analysis
Sonic 3.6 (27 August 2026) replaces Sonic 3.5 on this sheet. Cartesia reports 44 languages, a French blind preference of 86% against Eleven v3, and replies under 90 ms. Those are vendor figures. The public Speech Arena on 17 October 2026 lists Sonic 3.6 at ELO 1285, rank 4. The pricing FAQ the same day lists Pro overage at $65 per million credits. Ink-2 has spoken French since 17 September 2026; its 1.47% WER remains an English bench.
Strengths
- 44 langues, dont le français
- Sous 90 ms annoncé par Cartesia pour 3.6
- Meilleure tenue d’accent sur les clones, selon l’éditeur
- On-prem et endpoints localisés annoncés
- ELO public 1285, rang 4 (17 octobre)
Weaknesses
- Les 90 ms et le 86 % français sont des chiffres Cartesia
- Pas de timestamps lip-sync
- Ink-2 parle français depuis le 17 septembre 2026, mais le WER 1,47 % reste un banc anglais
- Résidence suisse non attestée
Voice Capabilities
Clonage instant. Sonic 3.6 annonce une meilleure tenue de l’accent, y compris sur des accents moins répandus.
Fine-grained control of emotion, speed and volume, suited to fast line-by-line direction.
How: `generation_config.emotion`, `speed`, `volume`, SSML tags and `[laughter]`.
Validate: Verify emotional coverage for the selected voice and language; the API does not replace upstream dramaturgical direction.
Cartesia annonce une réponse sous 90 ms pour Sonic 3.6. Le record 40 ms appartenait à la génération précédente et n’est pas repris comme mesure 3.6.
No native viseme/timestamp output. Requires external alignment.
Pricing
FAQ Cartesia du 17 octobre 2026 : dépassement Pro $65 / 1M crédits, Startup $45, Scale $38. 1 crédit = 1 caractère. L’ancien $46,70/1M n’est plus sur la page. 1 minute ≈ 750–800 crédits.
| Plan | Subscription/mo | Included | Overage/min | Max streams | Top-up |
|---|---|---|---|---|---|
Free No overage. 2 concurrent TTS requests. | Free | 26.667 min | — | 2 | — |
Pro Overage $65 / 1M credits. 3 concurrent TTS requests. | $5/mo | 133.333 min | $0.0487 | 3 | Yes |
Startup Overage $45 / 1M credits. 5 concurrent TTS requests. | $49/mo | 1,666.667 min | $0.0338 | 5 | Yes |
ScaleRecommended Overage $38 / 1M credits. 15 concurrent TTS requests. | $299/mo | 10,666.667 min | $0.0285 | 15 | Yes |
Enterprise Custom concurrent limits | Free | Custom | — | ∞ | Yes |
Sovereignty & Compliance
On-prem et cloud client annoncés. À contractualiser avant de parler de souveraineté suisse.
Data residency: US par défaut. Endpoints localisés et on-prem annoncés ; résidence suisse non attestée.
Cartesia Sonic 3.6 — Strategic Positioning
Beyond technical specs: where does this tool sit in the ecosystem, what are the risks and strategic implications for GamiWays?
Cartesia's State Space Model delivers a structural 40ms latency advantage that's hard to replicate — but its cloud-only stance creates a sovereignty blind spot for regulated European markets.
A. Strategic Positioning
Target customer: Developer / Real-time AI agent builder
Ultra-low latency TTS (40ms TTFA) via State Space Model architecture — purpose-built for real-time conversational AI agents.
B. Competitive Moat
- State Space Model (SSM): linear scaling vs quadratic transformers — structural latency advantage
- 40ms TTFA with expressive capabilities (laughter, emotions) — hard to replicate without SSM expertise
- SOC 2, HIPAA, GDPR compliance — enterprise-ready from day one
Vulnerability: No on-premise option. Price pressure from competitors like Smallest AI ($0.01/min vs $0.03/min). Open-source models catching up in quality.
E. Strategic Questions for GamiWays
Sovereignty fit
Cloud-only with no on-premise option. GDPR compliant but no Swiss/EU-specific data residency guarantee. Risk for GamiWays Phase 2.
Build vs. Buy
Buy for Phase 1 (best real-time latency). For Phase 2 sovereignty, evaluate Inworld TTS (on-premise) or open-source alternatives.
Lock-in risk
SSM architecture creates technical lock-in (hard to replicate). But API-based integration allows migration to alternatives if needed.
Roadmap alignment
Excellent for Phase 1 real-time agents. Problematic for Phase 2 sovereignty requirements without on-premise option.
This sheet's verification
Cartesia Sonic 3.6, 27 Aug 2026. Public Speech Arena, 17 Oct 2026: ELO 1285, rank 4.
Update note: Arena publique du 17 octobre 2026 : Sonic 3.6 ELO 1285, rang 4 (intervalle 3–5). Le prix $46,70/1M n’est pas modifié.
This date covers provider-specific prices, capabilities and notes. Comparative benchmarks follow the synchronization and methodology shown above.
ELO benchmarks / indices: Artificial Analysis · Last API sync : 5 October 2026