PREDROP.ai
UpdateAI Voice

ElevenLabs Conversational AI 2.0 hits sub-300ms latency

The new release closes the gap with human turn-taking — and the dev pricing dropped 40%.

$0.05/min (was $0.08)

Read the update

Affiliate link · we may earn a commission

ElevenLabs released v2 of their Conversational AI stack. The big number is end-to-end latency — they're claiming median 280ms for the full speech-to-speech roundtrip, which is past the threshold where conversations stop feeling laggy.

Why we picked it

If you're shipping a voice agent in production, latency is the user-experience moat. v2 is the first OSS-adjacent stack that meaningfully competes with OpenAI's realtime voice on response time.

What's new

  • 280ms median latency (was ~520ms in v1)
  • Native turn-detection — agent stops talking when interrupted
  • Pricing dropped from $0.08 to $0.05 per agent-minute
  • Function calling now supports parallel tool invocation

Watch out for

  • The latency claim is for their own infra; self-hosted is slower
  • Pricing only drops for new accounts; existing accounts grandfather to 30 days
  • Function-calling works with Claude and OpenAI but not Gemini yet

If you've been holding off on shipping a voice product because of latency, this is the moment.

Tagselevenlabsvoice-aiapi

$0.05/min (was $0.08)

Read the update