AssemblyAI Universal-3.6 Pro Realtime
universal-3-6-pro (29 sept. 2026) — 32 langues, 0,45 $/h, endpointing d’entités. WER fiche inchangé.
Comparative Scores
Architecture
Voice Agent API très pertinent pour GamiWays Phase 1 : pipeline STT+LLM+TTS en 1 WebSocket simplifie l'architecture. Tool calling permet d'intégrer un RAG sur la base de connaissances GamiWays. Pas de clonage vocal natif : intégrer ElevenLabs ou Cartesia via TTS externe pour la voix du GamiWays. Référence précision WER pour benchmarking.
Analysis
On 29 September 2026 AssemblyAI released Universal-3.6 Pro Realtime (`universal-3-6-pro`): 32 languages, $0.45/hr, and entity-aware endpointing. Its English voice-agent WER of 5.19% is a vendor protocol and is not the sheet’s comparative WER. Earlier context: AssemblyAI launched its Voice Agent API on April 29, 2026: a complete STT+LLM+TTS pipeline in one WebSocket connection at a flat $4.50/hr. Universal-3 Pro Streaming (u3-rt-pro) is its real-time STT model, with semantic and acoustic turn detection, native barge-in, and session resumption. JSON Schema tool calling supports a custom RAG stack (Pinecone, LlamaIndex, and others) through function calling; there is no native RAG, but external integration is complete. There is no native voice cloning: voices are predefined (18+ English and multilingual voices), though an external TTS such as ElevenLabs or Cartesia can provide a custom voice. LeMUR features cover transcription summarisation, Q&A and sentiment. Supports 99 languages, diarisation and word-level timestamps.
Strengths
- Voice Agent API : pipeline STT+LLM+TTS complet en 1 WebSocket, $4.50/hr flat
- Universal-3 Pro Streaming : turn detection sémantique+acoustique, barge-in natif
- Tool calling JSON Schema → RAG custom intégrable (Pinecone, LlamaIndex, etc.)
- Session resumption 30s, live config update mid-conversation
- 4.9% WER Universal-2 — meilleure précision cloud
- 99 langues, diarisation, LeMUR AI (résumé, Q&R, sentiment)
- GDPR, SOC 2 Type 2, ISO 27001, HIPAA, PCI DSS
Weaknesses
- Pas de clonage vocal natif — voix prédéfinies uniquement (TTS externe possible via intégration)
- Pas de RAG natif — nécessite tool calling + infrastructure RAG externe
- Cloud only — pas d'option on-premise, souveraineté limitée
- Voice Agent API : LLM intégré non personnalisable nativement (custom LLM via Retell AI possible)
- $4.50/hr Voice Agent API — coût élevé pour usage intensif
STT Capabilities
universal-3-6-pro : même endpoint streaming, nom de modèle à changer. Endpointing qui garde le tour ouvert sur une entité. Latence médiane d’endpoint 537 ms, chiffre AssemblyAI, identique à 3.5 Pro.
32+ languages
Pricing
universal-3-6-pro : 0,45 $/h selon AssemblyAI le 29 septembre 2026. Le 0,57 $/h Pipecat de juin et le Voice Agent à 4,50 $/h restent des repères antérieurs.
| Plan | Subscription/mo | Included | Overage/min | Top-up |
|---|---|---|---|---|
Free 5 new streams/min rate limit on free tier | Free | $50 free credit | $0.0025 | — |
Pay As You GoRecommended Unlimited concurrency — no stream cap documented | Free | None | $0.0025 | Yes |
Sovereignty & Compliance
Cloud only. No on-premise. EU data residency available (GDPR, SOC 2 Type 2, ISO 27001, HIPAA, PCI DSS).
Data residency: US (default). EU data residency disponible. SOC 2 Type 2, ISO 27001, HIPAA, PCI DSS.
Cloud only. No on-premise option. EU data residency available.
AssemblyAI Universal-3.6 Pro Realtime — Strategic Positioning
Beyond technical specs: where does this tool sit in the ecosystem, what are the risks and strategic implications for GamiWays?
AssemblyAI is the audio intelligence leader — #1 accuracy on Hugging Face leaderboard, 30% fewer hallucinations, full PII/diarization suite. EU Dublin data residency available but no on-premise limits Phase 2 sovereignty.
A. Strategic Positioning
Target customer: Developer / Enterprise — audio intelligence, voice agents, Fortune 500
Ranked #1 on Hugging Face Open ASR Leaderboard with Universal-3 Pro — 30% fewer hallucinations than competitors, full audio intelligence suite.
B. Competitive Moat
- #1 on Hugging Face Open ASR Leaderboard — Universal-3 Pro with 30% fewer hallucinations
- Full audio intelligence suite: diarization, PII redaction, content moderation — beyond transcription
- SOC 2 Type 2, PCI-DSS 4.0 Level 1, ISO 27001 in progress — enterprise compliance
Vulnerability: Open-source Whisper catching up in quality. High switching costs if deeply integrated. No full on-premise option.
E. Strategic Questions for GamiWays
Sovereignty fit
EU data residency in Dublin available. No full on-premise. Strong compliance certifications reduce regulatory risk.
Build vs. Buy
Buy for Phase 1 (best accuracy, audio intelligence suite). For Phase 2 sovereignty, evaluate Whisper self-hosted for basic transcription.
Lock-in risk
Developer-focused API creates integration dependency. Audio intelligence suite features increase switching costs.
Roadmap alignment
Good for Phase 1 voice agents and audio intelligence. Phase 2 sovereignty requires self-hosted alternatives for full data control.
This sheet's verification
AssemblyAI Voice Agent API launch (Apr 29, 2026) + pricing page
Update note: Universal-3.6 Pro Realtime vérifié le 30 septembre 2026 : modèle universal-3-6-pro, 32 langues, 0,45 $/h, endpoint 537 ms (éditeur). Le WER AA/Pipecat de la fiche n’est pas écrasé.
This date covers provider-specific prices, capabilities and notes. Comparative benchmarks follow the synchronization and methodology shown above.
ELO benchmarks / indices: Artificial Analysis