fix(models): keep OpenRouter :batch variants out of chat routing (#13622)

* fix(models): keep OpenRouter :batch variants out of chat routing

ModelSync imported OpenRouter's Batch-API-only variants into the chat
catalogue. A chat completion against one is rejected upstream with

  404 This model is only available through the Batch API.
      Use the /api/beta/batches endpoint instead.

which #13596 measured 91 times in 41 hours, third by volume, plus the
`model not found - locking mode` failover churn behind it.

OpenRouter's /models carries no endpoint metadata that separates a batch
variant from a chat one, so `classifyExplicitEndpoints` cannot decide it
and the rule belongs in `modelEndpointPolicy`, beside the OpenAI
image/video policy and for the same reason: the file exists so discovery,
import and catalog projection agree on one answer.

Matched as the exact `:batch` suffix, not "has a variant suffix" --
`:free`, `:nitro`, `:floor`, `:online`, `:extended` and `:thinking` are
routing hints on the same chat model, and excluding them would silently
shrink the routable catalogue. Applied unconditionally for this provider:
there is no "batch" endpoint name an upstream could declare next to a chat
one, and the already-stored rows carry the synthetic `["chat"]` default
that re-imported them in the first place.

Closes #13596

* docs(changelog): add fragment for the OpenRouter batch-variant fix

changelog.d/fixes/13622-openrouter-batch-variants.md

@@ -0,0 +1 @@
+- **fix(models):** keep OpenRouter's Batch-API-only `:batch` variants out of chat routing — ModelSync imported all 77 of them into the chat catalogue, where every request that landed on one was rejected with `404 This model is only available through the Batch API` (#13622)

open-sse/services/modelEndpointPolicy.ts

@@ -71,12 +71,44 @@ function classifyOpenAiModel(modelId: string): ModelEndpointDecision | null {
   return null;
 }
 
+/**
+ * OpenRouter appends a variant suffix to a base model id: `:free`, `:nitro`,
+ * `:floor`, `:online`, `:extended`, `:thinking`. Almost all of them are routing
+ * hints on the same chat model and stay chat-selectable. `:batch` is the
+ * exception -- it names the Batch-API-only variant, and a chat completion
+ * against it is rejected upstream with
+ * `404 This model is only available through the Batch API. Use the
+ * /api/beta/batches endpoint instead.` (issue #13596: 91 of those in 41 hours,
+ * plus the `model not found - locking mode` failover churn behind them).
+ *
+ * OpenRouter's `/models` carries no endpoint metadata that separates the two,
+ * so `classifyExplicitEndpoints` cannot decide it and the knowledge belongs
+ * here with the rest of the exceptional provider policy. Deliberately an exact
+ * suffix rather than "has a variant suffix": the other suffixes above are chat
+ * models and excluding them would silently shrink the routable catalogue.
+ */
+const OPENROUTER_BATCH_SUFFIX = ":batch";
+
+function classifyOpenRouterModel(modelId: string): ModelEndpointDecision | null {
+  return modelId.trim().toLowerCase().endsWith(OPENROUTER_BATCH_SUFFIX)
+    ? { kind: "non-chat", chatSelectable: false, reason: "provider-policy" }
+    : null;
+}
+
 export function getModelEndpointDecision(
   provider: string | null | undefined,
   modelId: string,
   supportedEndpoints?: readonly string[]
 ): ModelEndpointDecision {
   const explicit = classifyExplicitEndpoints(supportedEndpoints);
+  if (provider?.trim().toLowerCase() === "openrouter") {
+    // Unconditional, unlike the OpenAI branch below: there is no "batch"
+    // endpoint name an upstream could declare alongside a chat one, and the
+    // rows already stored for these carry the synthetic `["chat"]` default --
+    // letting that win would keep importing exactly the variants this excludes.
+    const openRouterDecision = classifyOpenRouterModel(modelId);
+    if (openRouterDecision) return openRouterDecision;
+  }
   if (provider?.trim().toLowerCase() === "openai") {
     const openAiDecision = classifyOpenAiModel(modelId);
     if (openAiDecision) {

tests/unit/model-endpoint-policy.test.ts

@@ -28,6 +28,65 @@ test("OpenAI video models are not chat-selectable without upstream endpoint meta
   });
 });
 
+test("OpenRouter :batch variants are not chat-selectable", () => {
+  for (const modelId of [
+    "google/gemini-3.6-flash:batch",
+    "anthropic/claude-sonnet-4.5:batch",
+    "minimax/minimax-m3:batch",
+    "inkling:batch",
+  ]) {
+    assert.deepEqual(getModelEndpointDecision("openrouter", modelId), {
+      kind: "non-chat",
+      chatSelectable: false,
+      reason: "provider-policy",
+    });
+  }
+});
+
+test("OpenRouter's other variant suffixes stay chat-selectable", () => {
+  // Excluding these would shrink the routable catalogue -- they are routing
+  // hints on the same chat model, not a different endpoint.
+  for (const modelId of [
+    "google/gemini-3.6-flash:free",
+    "anthropic/claude-sonnet-4.5:thinking",
+    "meta-llama/llama-4-70b:nitro",
+    "perplexity/sonar:online",
+    "google/gemini-3.6-flash",
+  ]) {
+    assert.equal(isChatSelectableModel("openrouter", { id: modelId }), true, modelId);
+  }
+});
+
+test("a synthetic chat default does not re-admit an OpenRouter batch variant", () => {
+  // The rows already stored for these carry `["chat"]` as the synthetic import
+  // default, which is exactly what re-imported them.
+  assert.equal(
+    isChatSelectableModel("openrouter", {
+      id: "google/gemini-3.6-flash:batch",
+      supportedEndpoints: ["chat"],
+    }),
+    false
+  );
+});
+
+test("the batch policy is scoped to OpenRouter", () => {
+  assert.equal(isChatSelectableModel("custom-provider", { id: "some-model:batch" }), true);
+  assert.equal(isChatSelectableModel(null, { id: "some-model:batch" }), true);
+});
+
+test("filterChatSelectableModels drops the batch variant and keeps its base model", () => {
+  const models = [
+    { id: "google/gemini-3.6-flash" },
+    { id: "google/gemini-3.6-flash:batch" },
+    { id: "google/gemini-3.6-flash:free" },
+  ];
+
+  assert.deepEqual(
+    filterChatSelectableModels("openrouter", models).map((model) => model.id),
+    ["google/gemini-3.6-flash", "google/gemini-3.6-flash:free"]
+  );
+});
+
 test("provider policy is scoped and does not classify another provider by model name", () => {
   assert.equal(
     isChatSelectableModel("custom-provider", { id: "gpt-image-shaped-chat-model" }),