Nox Voice 2 (New)

Voice model · Released in Nox: October 14, 2026 · Page last updated: October 17, 2026

The most powerful Nox voice yet — expressive, interruptible, Plan mode in the call

Nox Voice 2 is the new default voice in Nox and the most powerful voice model Nox has ever shipped — more than 50% less cost per call than Nox Voice 1 Advance. It is built to sound like a person on a call rather than a narrator: the model writes short delivery directions into its own reply — warmth, a pause, a quiet tone, a light chuckle — and the speech engine performs them. It replies in whatever language you just used, mixing languages mid-sentence when you do, and it stops the moment you say something.

Voice 2 is powered by Nox's own text models. In chat mode every turn is answered by Vela 2, the everyday Nox model, in a brief spoken style; in Plan mode you pick the model yourself on a five-level bar that runs from Vela Light up to Orion Extra, the flagship at deep effort. A separate expressive speech stage turns those answers into voice. Voice 2 is the third generation of spoken Nox: Nox Voice 1.2 brought real-time conversation, Nox Voice 1 Advance made voice an official, saved-to-chat part of Nox on every paid plan, and Voice 2 replaces the narrator-style delivery of both with a model that performs its own tone — at more than 50% less per call than Voice 1 Advance: speech synthesis costs 400 credits per 1,000 characters instead of 1,000 (−60%), transcription 65 credits per minute instead of 130 (−50%), and the text model is priced the same.

Voice 2 is a complete call pipeline. Your microphone audio passes through a speech gate that only forwards sound when it plausibly holds your voice, so silence and room noise are never transcribed or billed. Transcription runs live, the text model answers phrase by phrase, and synthesis starts on the first finished phrase, so speech begins while the rest of the answer is still being written.

What makes it expressive is that the model directs the delivery itself. Inside each reply it places short bracketed directions right before the words they colour — [warm], [soft, reassuring], [thinking out loud], [light chuckle], [short pause] — and the speech engine renders them as performance rather than reading them out. Directions are drawn from a fixed vocabulary of voice-only cues: no sound effects, no music, no accents, and never a playful tag on a dose, a medical fact, or emergency guidance.

Language follows you. Voice 2 answers in the language of your last sentence and, if you mixed two languages in one breath, mixes them the same way — switching mid-sentence, keeping drug names, numbers, and brand names in the language you said them in. The chat thread shows the plain text; the delivery directions never appear on screen.

You can interrupt it with words, not noise. While Nox is speaking the gate demands a clearly louder onset than the room, and the interruption only fires once recognised words arrive — a cough or a backchannel like “mm-hm” does not cut it off; “wait, what about…” does. Short stop phrases (“wait”, “hold on”, “等等”) stop playback immediately.

The call is not voice-only. You can attach photos mid-call — a label, a prescription, a rash — and Nox reads them before it answers; you can type instead of speaking at any point and the reply still comes back spoken; and a mute button holds your microphone without ending the call. Every completed exchange is saved into the same chat thread, so the conversation has a written record you can reread.

Plan mode runs inside the call. Switch to Plan and Voice 2 hands the turn to a text model chosen on a five-level bar — Vela Light, Vela Ultra, Orion Flash Balance, Orion Flash Max, and Orion Extra — with live web search and page reading always on (up to eight web steps per turn). The full written answer appears on screen in Markdown with links to the sources it actually read, while a short spoken version is read aloud; a server-written “what I did” line lists the real searches, pages, and credits used.

In chat mode, Voice 2 is answered by Vela 2, the everyday Nox model, with a brief-by-default spoken style: one to three sentences unless you ask for detail. Twelve voices are available and chosen in Settings; the choice applies to every Voice 2 call.

Voice 2 remembers your calls. When a call ends, Nox saves a short note of what mattered — medications, symptoms, decisions, open questions — and your next Voice 2 call can pick it up. This memory is built only from your own Voice 2 calls, never from your text chats, and is read only by Voice 2; it covers your last five calls within 30 days, is loaded while the call connects so it adds no delay to answers, and is deleted with your account.

Every turn runs behind Leo. Spoken input is screened by the same deterministic red-flag detector as text, an emergency is spoken in full before anything else, and the saved transcript carries the same safety banner the text chat would. Calls run for up to 15 minutes each and are available on every paid plan.

Capabilities

  • The most powerful Nox voice model yet, more than 50% less cost per call than Nox Voice 1 Advance
  • Powered by Vela 2 in chat; your pick of five Vela/Orion levels in Plan mode
  • Self-directed expressive delivery — the model places voice directions ([warm], [soft], [light chuckle], pauses) that the speech engine performs
  • Replies in the language you just spoke; mixes languages mid-sentence when you do
  • Interruption on recognised words, not noise — talking over Nox redirects the answer; stop phrases halt it instantly
  • Photos and typed text during a call, with spoken replies either way
  • Plan mode out loud: five-level text-model bar with live web search and page reading, written answer on screen, spoken summary aloud
  • Honest “what I did” line on every Plan turn — real searches, pages, and credits
  • Remembers what mattered in your recent Voice 2 calls — voice-only memory, never your text chats, loaded before you speak so answers stay fast
  • 12 selectable voices, chosen in Settings
  • Speech-gated transcription — silence and room noise are never transcribed or billed
  • Mute without hanging up; every completed exchange saved to the chat thread

Model details

  • Status — New — the default voice in Nox
  • Generation — Third-generation spoken Nox: Nox Voice 1.2 → Nox Voice 1 Advance → Nox Voice 2
  • Powered by — Vela 2 in chat mode; a five-level Vela/Orion bar in Plan mode; a separate expressive speech stage
  • Cost vs Voice 1 Advance — More than 50% less per call — speech 400 vs 1,000 credits per 1,000 characters (−60%), transcription 65 vs 130 credits per minute (−50%), same text-model price
  • Conversation loop — Speech-gated audio → live transcription → phrase-by-phrase answer → expressive synthesis from the first finished phrase
  • Delivery — Model-directed voice cues from a fixed, voice-only vocabulary; never shown in the transcript
  • Languages — Follows the language of your last sentence, including mid-sentence mixing
  • Interruptions — Yes — recognised words interrupt; stop phrases halt playback at once
  • Inputs during a call — Voice, typed text, and photos
  • Chat-mode model — Vela 2, brief by default (one to three spoken sentences)
  • Plan-mode models — Five levels: Vela Light, Vela Ultra, Orion Flash Balance, Orion Flash Max, Orion Extra
  • Plan-mode web — Always on — search and page reading, up to 8 web steps per turn
  • Plan-mode output — Written Markdown answer on screen + short spoken version + a “what I did” line
  • Memory — Recent Voice 2 calls only (last 5 within 30 days); never reads text chats; prefetched at connect, no added answer latency
  • Voices — 12, selected in Settings
  • Transcript — Yes — completed exchanges are saved to the chat thread
  • Per-call limit — Up to 15 minutes
  • Safety layer — Runs behind Leo's deterministic red-flag detector on every turn; emergencies are spoken in full first
  • Available on — Every paid plan (Pro and MAX)

Availability: Available on every paid plan (Pro and MAX). Not included on Free.

Nox Voice 2 — frequently asked questions

What is Nox Voice 2?

Nox Voice 2 is Nox's voice model. Nox Voice 2 is the new default voice in Nox and the most powerful voice model Nox has ever shipped — more than 50% less cost per call than Nox Voice 1 Advance. It is built to sound like a person on a call rather than a narrator: the model writes short delivery directions into its own reply — warmth, a pause, a quiet tone, a light chuckle — and the speech engine performs them. It replies in whatever language you just used, mixing languages mid-sentence when you do, and it stops the moment you say something.

What can Nox Voice 2 do?

Nox Voice 2's capabilities: The most powerful Nox voice model yet, more than 50% less cost per call than Nox Voice 1 Advance; Powered by Vela 2 in chat; your pick of five Vela/Orion levels in Plan mode; Self-directed expressive delivery — the model places voice directions ([warm], [soft], [light chuckle], pauses) that the speech engine performs; Replies in the language you just spoke; mixes languages mid-sentence when you do; Interruption on recognised words, not noise — talking over Nox redirects the answer; stop phrases halt it instantly; Photos and typed text during a call, with spoken replies either way; Plan mode out loud: five-level text-model bar with live web search and page reading, written answer on screen, spoken summary aloud; Honest “what I did” line on every Plan turn — real searches, pages, and credits; Remembers what mattered in your recent Voice 2 calls — voice-only memory, never your text chats, loaded before you speak so answers stay fast; 12 selectable voices, chosen in Settings; Speech-gated transcription — silence and room noise are never transcribed or billed; Mute without hanging up; every completed exchange saved to the chat thread.

Which Nox plans include Nox Voice 2?

Available on every paid plan (Pro and MAX). Not included on Free.

Related Nox models

Back to the Nox model family · Plans and pricing · Nox features