From a80f85692b8feec7f25c1459f0da0ab51632bd10 Mon Sep 17 00:00:00 2001
From: Abhimanyu Kapur <38531241+akapur99@users.noreply.github.com>
Date: Sat, 11 Jul 2026 16:43:36 -0700
Subject: [PATCH] fix(ui): drop max_tokens from the auto-router connection
 probe

max_tokens=1 makes reasoning models (o1/o3/...) return a 400 "max_tokens
reached" because reasoning tokens count against the cap, so a reachable
reasoning tier showed a false failure in Test Connection. Live-verified: o3
400s with the cap and succeeds without it.

Extract the request shape into a pure buildModelGroupTestRequest and cover it
with a test asserting the chat body carries no max_tokens (or
max_completion_tokens), so this regression is caught in unit tests instead of
only against a live reasoning model.
---
 .../src/components/networking.test.ts         | 16 +++++++++++++
 .../src/components/networking.tsx             | 24 ++++++++++++++-----
 2 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/ui/litellm-dashboard/src/components/networking.test.ts b/ui/litellm-dashboard/src/components/networking.test.ts
index 43f85cc8674..b9d04a00f61 100644
--- a/ui/litellm-dashboard/src/components/networking.test.ts
+++ b/ui/litellm-dashboard/src/components/networking.test.ts
@@ -514,3 +514,19 @@ describe("sessionSpendLogsCall", () => {
     expect(parsed.searchParams.get("page_size")).toBe("100");
   });
 });
+
+describe("buildModelGroupTestRequest", () => {
+  it("builds a chat completion request with NO max_tokens (reasoning models 400 on a tiny cap)", () => {
+    const { path, body } = Networking.buildModelGroupTestRequest("o3", "chat");
+    expect(path).toBe("/v1/chat/completions");
+    expect(body).toEqual({ model: "o3", messages: [{ role: "user", content: "test from litellm" }] });
+    expect(body).not.toHaveProperty("max_tokens");
+    expect(body).not.toHaveProperty("max_completion_tokens");
+  });
+
+  it("builds an embeddings request for embedding mode", () => {
+    const { path, body } = Networking.buildModelGroupTestRequest("text-embedding-3-small", "embedding");
+    expect(path).toBe("/v1/embeddings");
+    expect(body).toEqual({ model: "text-embedding-3-small", input: "test from litellm" });
+  });
+});
diff --git a/ui/litellm-dashboard/src/components/networking.tsx b/ui/litellm-dashboard/src/components/networking.tsx
index ae263727569..10d7f12604d 100644
--- a/ui/litellm-dashboard/src/components/networking.tsx
+++ b/ui/litellm-dashboard/src/components/networking.tsx
@@ -2324,17 +2324,29 @@ export type ModelGroupConnectionResult = { status: "success" } | { status: "erro
  * resolves the group, credentials, and provider. Used by the auto-router Test
  * Connection to probe each tier's model group and the embedding model.
  */
+/**
+ * Build the minimal request that probes a model group by public name. No
+ * max_tokens: reasoning models (o1/o3/...) reject a tiny cap with "max_tokens
+ * reached" because reasoning tokens count against it, which would show a false
+ * failure for a reachable tier.
+ */
+export const buildModelGroupTestRequest = (
+  modelGroup: string,
+  mode: "chat" | "embedding",
+): { path: string; body: Record<string, unknown> } =>
+  mode === "embedding"
+    ? { path: "/v1/embeddings", body: { model: modelGroup, input: "test from litellm" } }
+    : {
+        path: "/v1/chat/completions",
+        body: { model: modelGroup, messages: [{ role: "user", content: "test from litellm" }] },
+      };
+
 export const testModelGroupConnection = async (
   accessToken: string,
   modelGroup: string,
   mode: "chat" | "embedding",
 ): Promise<ModelGroupConnectionResult> => {
-  const path = mode === "embedding" ? "/v1/embeddings" : "/v1/chat/completions";
-  const body =
-    mode === "embedding"
-      ? { model: modelGroup, input: "test from litellm" }
-      : { model: modelGroup, messages: [{ role: "user", content: "test from litellm" }], max_tokens: 1 };
-
+  const { path, body } = buildModelGroupTestRequest(modelGroup, mode);
   try {
     await apiClient.post(path, { accessToken, body });
     return { status: "success" };