\begin{table}[htbp] \centering \caption{Model layer at the audited commit. The reference path targets the Anthropic Messages API; adaptive extended thinking is enabled only for the models listed as supporting it. OpenAI and Google models are served through an OpenAI-compatible adapter that receives the first 128 tools of the 245-tool catalogue. Model identifiers are reproduced verbatim from the client configuration (\texttt{client/src/lib/models.ts}); the Anthropic identifiers were checked against the provider's current model list, the others were not audited.} \label{tab:models} \begin{threeparttable} \footnotesize \begin{tabular}{@{}llccl@{}} \toprule Provider path & Model identifier & Adaptive thinking & Context & Role in selector \\ \midrule Anthropic (reference) & \texttt{claude-fable-5} & yes & 1M & default \\ Anthropic & \texttt{claude-opus-4-8} & yes & 1M & long-horizon agentic \\ Anthropic & \texttt{claude-opus-4-7} & yes & 1M & balanced \\ Anthropic & \texttt{claude-opus-4-6} & yes & 1M & fast and capable \\ Anthropic & \texttt{claude-sonnet-4-6} & yes & 1M & general purpose \\ Anthropic & \texttt{claude-haiku-4-5-20251001} & no & 200K & lowest latency \\ OpenAI-compatible & twelve OpenAI identifiers & n/a & -- & not audited \\ OpenAI-compatible & four Google identifiers & n/a & -- & not audited \\ \bottomrule \end{tabular} \begin{tablenotes}[flushleft] \footnotesize \item \textit{Notes.} Per-response generation is capped at 64{,}000 tokens and the temperature is set to~1 on the reference path; the summarisation call of Algorithm~\ref{alg:context} uses a cap of 2{,}000 tokens and a temperature of 0.5. The Beamer restructuring call (Section~\ref{subsec:docs}) is pinned to \texttt{claude-opus-4-8} irrespective of the selected model. Cost is computed from a single fixed schedule (\$3 per million input tokens, \$15 per million output tokens) for every model, so the ledger does not reflect per-model prices (Section~\ref{subsec:accounting}). \end{tablenotes} \end{threeparttable} \end{table}