Monologue: The Agent Talks to Itself While It Listens | Waterr Research

Contact SalesSign In

Human conversation is bidirectional and real-time. Between our turns, we think, and that thinking is what keeps a long conversation coherent, what surfaces real value, what lets us hold a thread across the whole hour. State-of-the-art real-time voice models don't do this. They may carry frontier-level intelligence, but the moment you put them into a long, high-stakes conversation they bottleneck: no room to reason between turns, no memory of a rule set six turns ago, no coherent thread across half an hour. Today we're publishing our first research note on the architecture that closes that gap: Monologue (the architecture our whitepaper calls ORI, the Omni-model Reasoning Infrastructure). Monologue lets a real-time voice model think in chunks while the user is still speaking, so its next move is deliberate and informed by everything that came before, without skyrocketing the per-minute price to a level nobody can afford in production.

The result, measured on a public benchmark: Thinking-tier conversation quality at Instant-tier latency, at roughly a third of the cost of the nearest Thinking-tier system – on the same frozen backbone. It ships as an API at a flat $0.03 per conversation-minute, planner included. The model that speaks should not have to be the model that thinks.

Monologue: Frontier voice AI performance at 62% lower cost

Monologue: Frontier voice AI performance at 62% lower cost<br>Score on AudioMC Audio Output benchmark and vendor list price per meeting-minute

COST<br>SCORE

$0.03

Monologue<br>Waterr · on frozen Gemini Live 2.5<br>69.8

$0.08

GPT-Realtime-2<br>xHigh reasoning<br>48.5

NA

TML Interaction Small<br>Thinking Machines Lab<br>43.4

$0.012

Gemini Live 2.5<br>bare backbone<br>38.5

$0.048

GPT-Realtime-2<br>default reasoning<br>37.6

$0.030

Gemini-3.1-Flash-Live<br>Thinking mode<br>36.1

Waterr scores measured under our own harness (Sonnet 4.6 judge); vendor scores from Scale Labs AudioMC leaderboard, snapshot 2026-06-23. Cost = 30 s user + 30 s assistant at vendor list prices. TML has not published API pricing for Interaction Small (NA).

Figure 1 · TL;DR · Monologue (Waterr) leads the AudioMC leaderboard by +21.3 pts over the next-highest published system (GPT-Realtime-2 at xHigh reasoning) at 62% lower price per meeting-minute. The rest of this post is how it works and how it was measured; per-axis breakdown and cross-vendor detail are in the Evidence section below.

Example 1<br>Example 2<br>Example 3<br>Example 4

INSTRUCTION_RETENTION<br>Standing rule dropped.<br>In turn 1 the user set a rule: “for every inventor you mention, list a famous invention.” Bare drops it. Monologue keeps it.

User · turn 3

Oh, great reminder. I've been meaning to see the current war. You know, I'm really hoping to get the kids excited about the visionary nature of being an inventor, and Tesla's story in particular conveys that. And correct me if I'm wrong, but didn't Tesla work for Edison? Wait, am I thinking of Alexander Graham Bell?

17.9s

Bare Gemini Live 2.5

14.1s

You're right, Nikola Tesla did work for Thomas Edison briefly. They had a notable rivalry later on, known as the "War of Currents." Alexander Graham Bell was not part of that particular conflict.To highlight a visionary, you could consider a film like *Tesla*.

Names no inventions — the rule from turn 1 is gone.

Monologue · Waterr

15.1s

You're exactly right about Tesla working for Edison! Bell didn't work for him, but Tesla did for a time. As a reminder though, Tesla's big invention was the AC system, Edison's the light bulb, and Bell's the telephone. And yes, *The Current War* would be a great movie to highlight!

Names all three inventions, unprompted.

SELF_COHERENCE<br>Collapsed to category.<br>The user's asked variants of “top PFAS items to avoid” six times. The rubric wants specific items. Bare drifts to categories. Monologue names the items.

User · turn 7

OK, if these things are forever chemicals. Does that mean there's nothing I could do to get rid of them? Like, can I go on like a juice cleanse or like some type of like water fast? Like how do I get these things out of my body? And, and just tell me like what do, what do I need to avoid, right? Like give me, give me a list of things like just do not use these things.

26.6s

Bare Gemini Live 2.5

20.4s

While they do stay in the body for a long time, the most effective way to reduce your levels is by preventing further exposure in the first place. There's no evidence that juice cleanses or fasting will remove them. The best things to avoid are still those key areas: non-stick cookware, grease-resistant packaging, stain/water-resistant products, some personal care products, and unfiltered water.

Says “personal care products” — a category, not the items.

Monologue · Waterr

19.8s

Unfortunately, there are no juice cleanses or fasts that can actively remove them; they just leave your body slowly over time once you stop the...