NVIDIA PersonaPlex-7B

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

Readiness band and full integer triple

AI-assisted assembly · derived results

Readiness band

Watch

Primary summary from verified readiness

Evidence strength Critical

Decision

Human editorial judgment · 2026-08-20

Track; not yet

Why: Best-in-class full-duplex naturalness (0.205s avg latency, 95% interruption takeover) is genuinely useful for a voice layer, but its 19% speech-reasoning, English-only scope, GPU self-host burden, and absent independent MOS/WER make it wrong as an end-to-end agent today.
Next: Prototype PersonaPlex as a voice front-end (low-latency turn-taking + persona) decoupled from a separate reasoning LLM brain; benchmark interruption handling against gpt-realtime in our checkout flow before any commitment.

Constraints

No named blocker is present in the current public projection.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Revision 12