Eleven v4 / v4 Turbo
Eleven v4 (28 sept. 2026) — 90+ langues, tags audio et Turbo ~150 ms premier son (éditeur). v3 reste le repère ELO d’août.
Comparative Scores
Architecture
Candidate for French performance direction on Où est Ava ? lines: audio tags and 10-second clones are the new facts to test. Keep v3 as the August quality baseline. Do not treat the vendor 150 ms or the claimed September rank as a measured end-to-end latency or as the portal ELO.
Analysis
Eleven v4 and v4 Turbo launched on 28 September 2026. The vendor describes 90+ languages, inline audio tags, natural-language direction, 10-second instant clones and a Turbo median time to first speech of about 150 ms. That latency is an ElevenLabs measurement with network removed. The portal ELO stays 1177 from the 18 August Artificial Analysis sync of Eleven v3; the September #1 claim is not copied into the ranking. v3 remains the dated quality reference. The character price on this sheet is still the May 2026 v3 list price.
Strengths
- Tags audio et direction en langage naturel (annonce v4)
- 90+ langues annoncées
- Clonage instant 10 s et Professional Voice Clone annoncés
- Turbo : ~150 ms de premier son médian (éditeur, réseau retiré)
- Timestamps mot à mot conservés sur l’API d’alignement
- v3 reste lisible comme repère ELO du 18 août (1177)
Weaknesses
- Le prix au caractère v4 n’est pas dans l’annonce ; le tarif fiche est encore celui de v3
- Les 150 ms sont un chiffre éditeur, pas une latence Où est Ava ?
- Cloud uniquement
- Le classement AA de septembre n’est pas resynchronisé
Voice Capabilities
v4 : clonage instant annoncé à partir de 10 secondes, plus Professional Voice Clone. Le palier v3 (zero-shot et affinage long) reste le repère antérieur.
A voice can receive performance directions and non-verbals in text; voice settings then stabilise the overall colour.
How: v4 audio tags ([laughs], natural-language direction, IPA); `stability` and `style` settings remain those of the previous generation.
Validate: v4 tags steer a performance: validate French reproducibility by voice and critical line. The 150 ms Turbo first-speech figure is a vendor number.
v4 Turbo : premier son médian ~150 ms et inférence médiane ~100 ms, chiffres ElevenLabs de septembre 2026, réseau retiré. Flash v2.5 (75 ms) et Turbo v2.5 restent les repères de la génération précédente.
Word-level timestamps via Alignment API. Phoneme-level available.
Pricing
Tarif fiche inchangé : v3 ~$100/1M chars (réduction du 7 mai 2026), Flash v2.5 ~$45/1M. L’annonce v4 du 28 septembre ne publie pas un nouveau prix au caractère ; elle confirme 3× de crédits Creator+ jusqu’au 12 octobre 2026.
| Plan | Subscription/mo | Included | Overage/min | Top-up |
|---|---|---|---|---|
Free Non-commercial use only | Free | 10 min | $0.3600 | — |
Starter Overage billed at $0.20/min | $6/mo | 30 min | $0.2000 | Yes |
Creator Overage at $0.18/min | $22/mo | 121 min | $0.1800 | Yes |
ProRecommended Overage at $0.17/min | $99/mo | 600 min | $0.1700 | Yes |
Scale Overage at $0.17/min | $299/mo | 1,800 min | $0.1700 | Yes |
Business Low-latency TTS from $0.05/min. High concurrency on Enterprise. | $990/mo | 6,000 min | $0.1700 | Yes |
Sovereignty & Compliance
No on-premise. Enterprise VPC available on request.
Data residency: US (default). EU data residency on enterprise plan.
LiveAvatar Custom/LITE Spike (real-time)
An interactive real-time test will land here: ElevenLabs produces audio while LiveAvatar Custom/LITE handles synchronized video through LiveKit or WebRTC. Target pipeline for Où est Ava ? and Dilemme Plastique — see also the HeyGen LiveAvatar page.
LiveAvatar sandbox session + ElevenLabs synthesis — enabled once keys are configured on Manus.
Planned pipeline
- Start a LiveAvatar Custom/LITE session (sandbox)
- Display avatar video in the browser (LiveKit)
- Synthesize text via ElevenLabs pcm_24000 with real-time lip-sync
Manus ops guide: docs/manus-liveavatar-spike-keys.md
Eleven v4 / v4 Turbo — Strategic Positioning
Beyond technical specs: where does this tool sit in the ecosystem, what are the risks and strategic implications for GamiWays?
ElevenLabs is the $11B voice AI leader going off-cloud — its on-premise/on-device move (H1 2026) signals a strategic pivot to capture regulated enterprise markets previously locked out by sovereignty concerns.
A. Strategic Positioning
Target customer: Enterprise / Creator / Developer
Industry-leading voice quality and emotional range. Going off-cloud with on-premise and on-device options (early access H1 2026), targeting the full infrastructure spectrum.
B. Competitive Moat
- ELO top-3 quality with superior emotional nuance and audio markup tags
- Complete enterprise platform: 380+ voices, 70+ languages, deep integrations
- Series D $500M (Feb 2026) — $11B valuation, $330M+ ARR — massive R&D firepower
Vulnerability: Perceived high pricing (20× more expensive than Inworld) and growing open-source competition closing the quality gap.
E. Strategic Questions for GamiWays
Sovereignty fit
On-premise option coming H1 2026, but currently cloud-only. EU data residency available on enterprise plan.
Build vs. Buy
Buy for Phase 1 MVP (best quality, fast deployment). Reassess for Phase 2 when on-premise matures — or switch to Inworld/open-source for sovereignty.
Lock-in risk
Proprietary API with high switching costs due to quality gap. On-premise option reduces lock-in for Phase 2.
Roadmap alignment
Strong for Phase 1 (voice quality, cloning). Phase 2 sovereignty depends on on-premise maturity (H2 2026 at earliest).
This sheet's verification
Eleven v4 announcement, 28 Sep 2026. Portal ELO remains the 18 Aug 2026 Artificial Analysis sync of Eleven v3 (1177).
Update note: Eleven v4 / v4 Turbo vérifiés le 30 septembre 2026 sur l’annonce officielle. ELO 1177 conservé : c’est le snapshot v3 du 18 août, pas le rang de septembre.
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 : 18 August 2026