mirror of
https://github.com/hanzoai/papers.git
synced 2026-11-02 13:32:30 +00:00
55 lines
2.6 KiB
TeX
55 lines
2.6 KiB
TeX
\section{Comparison to Existing Platforms}
|
|
|
|
\label{sec:comparison}
|
|
% ============================================================================
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{lccccc}
|
|
\toprule
|
|
\textbf{Feature} & \textbf{Hanzo} & \textbf{OpenAI} & \textbf{Anthropic} &
|
|
\textbf{Replicate} & \textbf{Together} \\
|
|
\midrule
|
|
Confidential inference & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
FHE support & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
Agent capsules & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
Verifiable receipts & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
Multi-provider & \checkmark & $\times$ & $\times$ & \checkmark & \checkmark \\
|
|
Open model registry & \checkmark & $\times$ & $\times$ & \checkmark & \checkmark \\
|
|
Edge inference & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
On-chain SLA & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
Self-hostable & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
User-controlled keys & \checkmark & $\times$ & $\times$ & $\times$ & $\times$ \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Platform comparison. Hanzo AI Cloud is the only platform offering
|
|
confidential inference with verifiable attestation.}
|
|
\label{tab:comparison}
|
|
\end{table}
|
|
|
|
\paragraph{OpenAI.} Market leader in model quality. No confidential inference.
|
|
No verifiable outputs. No self-hosting. Data retention policy permits 30-day
|
|
logging. Closed-source models with no weight access.
|
|
|
|
\paragraph{Anthropic.} Strong safety focus. No confidential inference. No
|
|
agent isolation beyond conversation-level context windows. No self-hosting.
|
|
Constitutional AI provides behavioral guarantees, not cryptographic ones.
|
|
|
|
\paragraph{Replicate.} Open model marketplace. No confidential inference.
|
|
Shared GPU instances with no tenant isolation. No verifiable outputs. Good
|
|
developer experience for model deployment, but no sovereignty guarantees.
|
|
|
|
\paragraph{Together.} Open model serving with competitive pricing. No
|
|
confidential inference. No agent hosting. No verifiable outputs. Together
|
|
is a cheaper alternative to OpenAI for open models, not a fundamentally
|
|
different architecture.
|
|
|
|
\paragraph{Hanzo differentiation.} Hanzo AI Cloud is the only platform where
|
|
the provider mathematically cannot see the user's data (FHE), the agent's
|
|
memory is cryptographically private (capsule vault), and every inference result
|
|
is provably attributable (A-Chain receipts). This is not a feature advantage.
|
|
It is an architectural category difference.
|
|
|
|
% ============================================================================
|