mirror of
https://github.com/hanzoai/papers.git
synced 2026-11-03 02:27:30 +00:00
61 lines
2.2 KiB
TeX
61 lines
2.2 KiB
TeX
\section{Zen Model Economics}
|
|
|
|
\label{sec:zen}
|
|
% ============================================================================
|
|
|
|
The Zen model family comprises 26 specialized models built on the Qwen3+
|
|
architecture. Each model has distinct compute costs and pricing:
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{llrr}
|
|
\toprule
|
|
\textbf{Model} & \textbf{Specialization} & \textbf{Params} &
|
|
\textbf{Cost/1K tokens} \\
|
|
\midrule
|
|
zen-coder & Code generation & 72B & 0.12 HANZO \\
|
|
zen-artist & Image generation & 14B & 0.45 HANZO \\
|
|
zen-scribe & Long-form writing & 32B & 0.08 HANZO \\
|
|
zen-designer & UI/UX generation & 14B & 0.35 HANZO \\
|
|
zen-analyst & Data analysis & 32B & 0.10 HANZO \\
|
|
zen-reason & Chain-of-thought & 72B & 0.15 HANZO \\
|
|
zen-chat & Conversational & 8B & 0.03 HANZO \\
|
|
zen-embed & Embedding & 1.5B & 0.01 HANZO \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Selected Zen models with per-1K-token pricing in HANZO. Full catalog of 26 models available at docs.hanzo.ai.}
|
|
\label{tab:zen}
|
|
\end{table}
|
|
|
|
\subsection{Model Provider Economics}
|
|
|
|
Third-party model providers can list models on the Hanzo Marketplace. The
|
|
economic structure for providers:
|
|
|
|
\begin{itemize}[leftmargin=1.1em]
|
|
\item \textbf{Listing fee}: One-time 1,000 HANZO to list a model (burned).
|
|
\item \textbf{Revenue split}: Provider receives 70\%, platform receives 30\%.
|
|
\item \textbf{Quality bond}: Providers stake 10,000 HANZO per model, slashable
|
|
for SLA violations.
|
|
\item \textbf{Governance approval}: New models require DAO vote for inclusion
|
|
in the default catalog.
|
|
\end{itemize}
|
|
|
|
\subsection{Training-Free Adaptation}
|
|
|
|
Hanzo's Active Semantic Optimization (ASO) and Decentralized Semantic
|
|
Optimization (DSO) protocols~\cite{hanzoaso, hanzodso} enable zero-training model
|
|
adaptation via Product-of-Experts (PoE) decoding. Users pay for experiential
|
|
prior retrieval at a fraction of fine-tuning cost:
|
|
|
|
\begin{itemize}[leftmargin=1.1em]
|
|
\item Prior retrieval: 0.001 HANZO per query (10$\times$ cheaper than
|
|
inference).
|
|
\item Prior submission: Free, but requires quality bond of 25 HANZO.
|
|
\item Quality rewards: High-quality priors earn 0.01 HANZO per downstream
|
|
usage.
|
|
\end{itemize}
|
|
|
|
% ============================================================================
|