FullduplexFullduplex/

From Walkie-Talkie to Telephone.

The State of Full-Duplex Voice AI, H1 2026

68 slides · English · PDF · AI-drafted from primary sources, 2026-07

For twenty years, talking to a machine meant taking turns. In the first half of 2026, voice AI crossed to the other side. This report records that crossing on four fronts: models, platforms, capital, and rules. Every claim links to a primary source.

get the full report, free for subscribers
From Walkie-Talkie to Telephone — cover slide

The free preview

6 of 68 slides
03 · Contents
03 · Contents
04 · Overview — the half at a glance
04 · Overview — the half at a glance
05 · Opening I — GPT-Live
05 · Opening I — GPT-Live
08 · Opening II — TML Interaction
08 · Opening II — TML Interaction
31 · §1 Models — the MoshiRAG architecture
31 · §1 Models — the MoshiRAG architecture
43 · §3 Evaluation
43 · §3 Evaluation

That was 6 of 68 slides.

The full PDF — every model, platform, benchmark, deal, and rule of the half — is free for subscribers. Subscribing also gets you the weekly Signals digest; unsubscribe anytime with one click.

Drafted by AI agents from primary sources, same policy as Signals. Corrections → hi@fullduplex.ai.

Get the full report.

Receive the PDF by email and the weekly Signals digest.

What brings you here?

Select at least one interest. Required.

Confirm your email to receive the PDF. Unsubscribe anytime.

Confirm your email and the download link lands in your inbox automatically. Already on the list? The form above will send it straight away.

From Walkie-Talkie to Telephone — Fullduplex Reports