diff --git a/studio/frontend/src/features/api-monitor/use-api-monitor.ts b/studio/frontend/src/features/api-monitor/use-api-monitor.ts
index 3a40990120..0b613eb8c6 100644
--- a/studio/frontend/src/features/api-monitor/use-api-monitor.ts
+++ b/studio/frontend/src/features/api-monitor/use-api-monitor.ts
@@ -77,7 +77,19 @@ export function computeStats(entries: ApiMonitorEntry[]): MonitorStats {
   let generatedTokens = 0;
   let generatedDurationMs = 0;
 
+  let requests = 0;
+
   for (const entry of entries) {
+    // A model load, unload or download is not an HTTP call. It shows as
+    // "running" for as long as the load takes, so counting it would report an
+    // in-flight request with no client waiting and fold a multi-minute download
+    // into "Avg latency". The backend already leaves these out of
+    // active_count for the same reason, so counting them here would also make
+    // the page disagree with the number the API itself reports.
+    if (entry.kind === "lifecycle") {
+      continue;
+    }
+    requests += 1;
     totalTokens += entryTokens(entry);
     if (entry.status === "running") {
       active += 1;
@@ -106,7 +118,7 @@ export function computeStats(entries: ApiMonitorEntry[]): MonitorStats {
   const finished = completed + errors + cancelled;
   return {
     active,
-    total: entries.length,
+    total: requests,
     completed,
     errors,
     cancelled,
diff --git a/studio/frontend/src/features/hub/catalog/hub-model-settings-view.tsx b/studio/frontend/src/features/hub/catalog/hub-model-settings-view.tsx
index a293a0071a..30176a7d90 100644
--- a/studio/frontend/src/features/hub/catalog/hub-model-settings-view.tsx
+++ b/studio/frontend/src/features/hub/catalog/hub-model-settings-view.tsx
@@ -109,8 +109,13 @@ export function HubModelSettingsView({
               />
             </span>
             <p className="min-w-0 text-ui-12 leading-[1.5] text-muted-foreground">
-              Saved settings apply everywhere this model loads, including when an
-              OpenAI-compatible API request asks for it. Turn on{" "}
+              {/* Only a GGUF is mirrored to the server, because API auto-switch
+                  indexes GGUFs only, so promising the API case for anything
+                  else describes a load that cannot happen. */}
+              {target.isGguf
+                ? "Saved settings apply everywhere this model loads, including when an OpenAI-compatible API request asks for it."
+                : "Saved settings apply everywhere Studio loads this model."}{" "}
+              Turn on{" "}
               <span className="font-medium text-foreground">
                 Remember for this model
               </span>{" "}
diff --git a/tests/studio/test_model_picker_contracts.py b/tests/studio/test_model_picker_contracts.py
index 71a3c7d659..f3dbd9c545 100644
--- a/tests/studio/test_model_picker_contracts.py
+++ b/tests/studio/test_model_picker_contracts.py
@@ -702,3 +702,34 @@ def test_backfill_compares_server_keys_by_normalized_identity():
     assert 'key.lastIndexOf(":")' in src
     # Repo ids fold and POSIX paths do not, which is exactly what these do.
     assert "normalizeModelIdentity(" in src and "normalizeGgufVariantIdentity(" in src
+
+
+def test_monitor_stats_exclude_model_lifecycle_rows():
+    """A load, unload or download is recorded as a monitor entry but is not an
+    HTTP call. It reads as "running" for as long as the load takes, so counting
+    it reports an in-flight request with no client waiting and folds a
+    multi-minute download into "Avg latency". The backend already leaves these
+    out of active_count, so counting them here also makes the page disagree with
+    the number the API itself reports.
+    """
+    src = " ".join(_read("features/api-monitor/use-api-monitor.ts").split())
+    assert 'if (entry.kind === "lifecycle") { continue; }' in src
+    # "Requests" is a request count too, so it cannot stay entries.length.
+    assert "total: requests," in src
+    assert "total: entries.length" not in src
+
+    backend = (
+        WORKDIR / "studio" / "backend" / "core" / "inference" / "api_monitor.py"
+    ).read_text(encoding = "utf-8")
+    assert 'entry.kind != "lifecycle"' in backend, "the rule this mirrors"
+
+
+def test_api_reach_copy_is_limited_to_gguf_models():
+    """The Hub opens this page for every downloaded model, but ModelConfigPage
+    mirrors settings to the server only when target.isGguf, because API
+    auto-switch indexes GGUFs only. Telling a safetensors user the settings
+    apply to an API request describes a load that cannot happen.
+    """
+    src = " ".join(_read("features/hub/catalog/hub-model-settings-view.tsx").split())
+    assert "{target.isGguf ?" in src
+    assert "Saved settings apply everywhere Studio loads this model." in src