ZZAZZ

A new class of
foundation model.

The Large Pricing Model.

Trained on transactions. Judged by what the market paid. LLMs predict the next token. The LPM predicts the next price.

Live quote $1.22 Avg. confidence 72.6%
<700ms
End-to-end
72.6%
Accuracy
14
Trust factors
32+
Genome attrs
Inside the architecture
What is the LPM

A new architecture,
purpose-built for value.

LLMs predict the next token. The LPM predicts the next price. Same mathematical core. Different objective. Language models are trained on text and judged by humans; the LPM is trained on transactions and judged by the market.

01 · Benchmark
QMV
Quantitative Market Value
The reference price. Free, public, auditable. Computed from content and comparables, with environment masked.
Public · on the Index
02 · For humans
QAP
Quantitative Adaptive Price
The execution price. Full state tensor — content, publisher, user, event. Computed on-site, in the publisher's audience context. What a reader pays in cash or attention.
Bounded · live · settles transactions
03 · For machines
QAP++
AI Execution Price
Per-token metered access for AI agents — training, RAG, summarization. Powered by CARL — Complex Adaptive Reinforcement Learning. Rights-aware. Audit-logged. Settles atomically.
AI agents · per-token · audited
04 · Enterprise
QMVprivate
Quantitative Market Value · Private
A private LPM instance for the enterprise. Trained on the firm's own corpus, same architecture as public QMV. The reference price, held inside the firm.
Walled · audited · on-premise

Generation is the easier problem — the model produces text; whether it's good is a human judgment. Pricing is harder. The model produces a number, and whether that number is right is decided by the market — by whether someone pays for it. There is a settlement.

In production since 2024 · 300+ patents · Trained on real settlements
The exchange

Live.
Right now.

Every piece of content the network knows about — being repriced. Some refreshs every two seconds, others once a minute. Each at its own frequency.

Exchange · Live
10,936,481 items priced · 11 visible · avg quote $1.89
Title · Brand Event Impact Quote LPM Freq Demand Change
The premise

Value is a function of supply,
demand, and time.

The function is non-monotonic. The same article can be worth $0.10 on Tuesday and $40 on Wednesday.

Three asset classes · three pricing models
1859
Oil
Spot pricing emerged.
Crude became a tradeable commodity.
~100M barrels priced daily
1973
Options
Black-Scholes published.
The CBOE opened.
~50M contracts priced daily
2024
Information
The Large Pricing Model went live.
Content became a priced asset.
34M items priced per second

Black-Scholes turned options into an asset class. The Large Pricing Model turns information into one. The underlying changes — from a contract on a stock to a sentence, an article, a video, a dataset. The principle holds: a price, marked continuously to live supply and demand.

One pass · looping

One model. One pass.
Looping forever.

State tensor in. Four genomes decompose. The policy network fires across 300 to 30,000 comparables. The model outputs a bounded price. Then it runs again — every 0.8 to 1.2 seconds.

EU Passes Landmark AI Regulation
5d ago · ft.com · Politics
Quote
$1.22
Frequency
1.20s
Confidence
72.6%
This single piece of content will be repriced 847 times in the next fifteen minutes.
Policy
QMV
QAP soon
QAP++ soon
Structural Book Value — masks live demand signals. Stable benchmark pricing.
↻ new quote every 0.8–1.2s
↑ next quote
St = Concat( xcontent, xpublisher, xuser, xeconomic ) → State Tensor
Content
95+ params · what is it
Publisher
40+ · who made it
User
23 · who wants it
Economic
ε_event · what is happening
1.14s
Genome vectors ↓
HNSW search → 17,551 comparables found
17,551
comparables
1.14s
QMV / QAP → feeds back into next comparison cycle →
17,551
Comparables
5.05M
Neurons
847
Guardrails
14
Trust factors
Frequency
Demand
EU_AI_REG 1m · QMV LIVE
$1.435
+24.6%
Open$1.440
High$1.445
Low$1.430
Close$1.435
Comparables17,551
Demand185,000
Quote #80
Live · supply & demand
Two signals.
Read every second.
Supply
comparable items live
Demand
attention pointed at it
Per content · demand elasticity
How willingness to pay bends
with price.
Breaking Recent Settled
Forecast

The next price —
and the prices that follow.

Each quote includes a forward forecast with an explicit, empirically calibrated confidence interval. The interval widens with horizon — a few basis points at the next tick, several percent at five minutes.

01
Variance scales with horizon
Each additional tick admits more exogenous shocks. The realized-error distribution widens monotonically with forecast distance.
02
Empirical calibration
The 95% interval is fit to the realized-error distribution of past prediction-settlement pairs. Recomputed continuously as new settlements arrive.
Trust

Fourteen signals.
Three pillars.

The reasoning is exposed. The weights stay sealed. Every quote ships with the evidence behind it — the comparables it leaned on, the reliability of the model state, the events shaping the moment.

Avg. confidence
72.64%
14 trust factors · live
< 45% · Low
45–85% · Mid
85%+ · High
Comparability
(6)
How well the model anchors this quote to similar content already priced.
Similar articles17,551
Match quality94.7%
Benchmark$1.20–$1.22
Content similarity91.3%
Publisher overlap67.8%
Topic coherence88.4%
Reliability
(5)
How stable and well-fed the model is at the moment of pricing.
Frequency1.20s
Neurons activated5.05M
Stabilityhigh
Data freshness98.2%
Model agreement96.1%
Explainability
(3)
What's pushing the price up or down, and how certain the model is about it.
Avg. confidence72.64%
Event impact+16%
Time decay−14%

An LLM tells you what the world might say.
The LPM tells you what the world is willing to pay.

At scale

That was one item.
There are millions.

Each with its own genome, its own market behavior, its own pricing curve. The LPM runs continuously across the network — quoting, anchoring, repricing. Trillions of interactions, every day.

34M stories & videos · priced per second
34M
Stories & videos
priced per second
105TB+
Daily data
150+
Languages

The weights tune themselves.
No human touches them.
At this speed, none could.

ZZAZZ
Large Pricing Model