Deepgram ships Aura 2 with sub-200ms latency for production voice agents

Read the original article →

What happened

Deepgram released Aura 2 on July 23, a text-to-speech model optimized for voice agents with sub-200ms time-to-first-byte latency. The model supports 40 voices across 8 languages with emotion and pacing controls.

Deepgram said Aura 2 powers more than 50 million agent minutes monthly across healthcare and customer-service deployments. Pricing starts at $0.015 per 1,000 characters.

Why it matters

Sub-200ms latency is the threshold where voice agents feel conversational rather than turn-based. That matters for support calls and intake flows where pauses break trust.

Deepgram is competing directly with ElevenLabs and xAI on the agent infrastructure layer, not just transcription.

MintedBrain take

Latency benchmarks vary by region and network path. Test Aura 2 from your deployment geography on real call flows before committing to a voice stack.

References

This article was originally published at VentureBeat. For the full piece, read the original article.

Discussion

  • Loading…

← Back to News