diff --git a/changelog.d/fixes/12374-combo-exclusion-reason-output-tokens.md b/changelog.d/fixes/12374-combo-exclusion-reason-output-tokens.md
new file mode 100644
index 0000000000..d89a27bfec
--- /dev/null
+++ b/changelog.d/fixes/12374-combo-exclusion-reason-output-tokens.md
@@ -0,0 +1 @@
+- **fix(combo):** capability-filter exhaustion caused by `max_tokens` above every target's known output limit now reports that reason (requested `max_tokens` vs the pool's highest known ceiling) instead of the unrelated "supports structured output" message ([#12229](https://github.com/diegosouzapw/OmniRoute/issues/12229)) — thanks @DW-MediaLab
diff --git a/open-sse/services/combo/comboStructure.ts b/open-sse/services/combo/comboStructure.ts
index 8d7adc068a..137561fbaa 100644
--- a/open-sse/services/combo/comboStructure.ts
+++ b/open-sse/services/combo/comboStructure.ts
@@ -627,6 +627,18 @@ export type CompatFilterOptions = {
   failOpen?: boolean;
 };
 
+function highestKnownOutputLimit(targets: ResolvedComboTarget[]): number {
+  let ceiling = 0;
+  for (const target of targets) {
+    const limit = getResolvedModelCapabilities({
+      provider: target.providerId || target.provider || null,
+      model: target.modelStr,
+    }).maxOutputTokens;
+    if (typeof limit === "number" && limit > ceiling) ceiling = limit;
+  }
+  return ceiling;
+}
+
 export function hasHardCapabilityFailure(reasons: string[]): boolean {
   return reasons.some((reason) => HARD_COMPAT_REASONS.has(reason));
 }
@@ -668,6 +680,16 @@ export function describeCapabilityFilterExhaustion(
     message = `No target in combo ${name} supports tool calling; request carried ${toolCount} tools`;
   } else if (primary === "vision") {
     message = `No target in combo ${name} has confirmed vision support for this image request`;
+  } else if (primary === "output_tokens") {
+    // #12229: name the real reason. Collapsing this into the structured-output
+    // message sent operators chasing response_format when the request's
+    // max_tokens simply exceeded every target's known output ceiling.
+    const ceiling = highestKnownOutputLimit(
+      rejected.filter((entry) => entry.reasons.includes("output_tokens")).map((e) => e.target)
+    );
+    message =
+      `No target in combo ${name} can produce the requested max_tokens=${requirements.requestedOutputTokens}; ` +
+      `the highest known output limit in the pool is ${ceiling}`;
   } else {
     message = `No target in combo ${name} supports structured output for this request`;
   }
diff --git a/tests/unit/8488-capability-filter-fail-closed.test.ts b/tests/unit/8488-capability-filter-fail-closed.test.ts
index 2be07aa7e7..418d9be453 100644
--- a/tests/unit/8488-capability-filter-fail-closed.test.ts
+++ b/tests/unit/8488-capability-filter-fail-closed.test.ts
@@ -320,3 +320,32 @@ test("auto context estimate still dispatches when all known limits look too smal
   assert.equal(result.status, 200);
   assert.deepEqual(dispatches, ["openai/tiny"]);
 });
+
+test("#12229 exhaustion: output_tokens exclusion names max_tokens vs the model ceiling", () => {
+  saveModelsDevCapabilities({
+    claude: {
+      "claude-haiku-4-5-20251001": capabilityEntry(200000, {
+        tool_call: true,
+        structured_output: true,
+        limit_output: 64000,
+      }),
+    },
+  });
+
+  const targets = [target("claude", "claude/claude-haiku-4-5-20251001")];
+  const body = {
+    messages: [{ role: "user", content: "hoi wie ben je?" }],
+    max_tokens: 100000,
+  };
+
+  const exhaustion = describeCapabilityFilterExhaustion(targets, body, "hermes-main");
+  assert.ok(exhaustion);
+  assert.deepEqual(exhaustion!.unmet, ["output_tokens"]);
+  assert.equal(exhaustion!.excluded[0].reason, "output_tokens");
+  assert.equal(
+    exhaustion!.message,
+    "No target in combo hermes-main can produce the requested max_tokens=100000; the highest known output limit in the pool is 64000"
+  );
+  assert.doesNotMatch(exhaustion!.message, /structured output/i);
+  assert.equal(exhaustion!.terminalReason, "capability_mismatch");
+});