diff --git a/hw/straylight/default.nix b/hw/straylight/default.nix
index 81120f2..39739fd 100644
--- a/hw/straylight/default.nix
+++ b/hw/straylight/default.nix
@@ -44,6 +44,14 @@ let
     model-draft = ${llamaModelsDir}/gemma4-26b-a4b-abliterated/mtp-gemma-4-26B-A4B-it.gguf
     spec-type = draft-mtp
 
+    ; Arcee Trinity Mini (26B total / 3B active MoE, US-trained "afmoe" architecture).
+    ; Benchmark candidate C — the US-built MoE, positioned as a reasoning/supervisor
+    ; model. Native 128k context (afmoe.context_length = 131072; the [*] 262144/2 split
+    ; lands exactly there). Local-file load like the other benchmark candidates; fetched
+    ; by download.sh, not --hf-repo.
+    [trinity-mini]
+    model = ${llamaModelsDir}/trinity-mini/Trinity-Mini-Q4_K_M.gguf
+
     ; Poolside Laguna S 2.1 (118B total, 8B active): primary coding agent.
     ; ~69 GB of weights, so it only loads once the raised TTM/GTT limit is active (reboot).
     ; DFlash GGUF is on disk (laguna-s-2.1-DFlash-BF16.gguf). nixpkgs llama.cpp 0.4.0